PROJECT

MANAGEMENT
with CPM, PERT and
Precedence
Diaarammina w V

THIRD EDITION

Joseph J. Moder
Professor, Department of Management Science and Computer
Information Systems, University of Miami

Cecil R. Phillips
Vice President, Kurt Salmon Associates, Inc.

Edward W. Davis
Professor, Colgate Darden Graduate School of Business
Administration, University oJ Virginia

VAN NOSTRAND REINHOLD COMPANY

New York
Copyright O 1983 by Van Nostrand Reinhold Company Inc.
Library of Congress Catalog Card Number: 82-16035
ISBN: 0-442-25415-6
All rights reserved. Certain portions of this work 6 1970, 1964 under the title
Project Management with CPM and PERT by Van Nostrand Reinhold Company
Inc. No part of this work covered by the copyright heronmay be reproduced or
used in any form or by any means-graphic, electronic, or mechanical, including
photocopying, recording, taping, or information storage and retrieval systems-
without permission of the publisher.
Manufactured in the United States of America
Published by Van Nostrand Reinhold Company Inc.
135 West 50th Street
New York, New York 10020
Van Nostrand Reinhold Company Limited
Molly MiUars Lane
Wokingham, Berkshire R G l l 2PY, England

Van Nostrand Reinhold

480 Latrobe Street
Melbourne, Victoria 3000, Australia

Macmillan of Canada
Division of Gage Publishing Limited
164 Commander Boulevard
Agincourt, Ontario MIS 3C7, Canada

15 14 13 12 11 10 9 8 7 6 5 4

Library of Cmgrers Catalogingin Publication Data

Moder, Joseph J.
Project management with CPM, PERT, and precedence
diagramming.
Rev. ed. of: Project management with CPM and PERT.
2nd ed. 1970.
Includes bioliographical references and index.
1. Industrial project management. 2. Critical path
analysis. 3. PERT (Network analysis) I. Phillips,
Cecil R. 11. Davis, Edward Willmore, 1941-
111. Title. IV. Title: Project management with C.P.M.,
P.E.R.T., and precedence diagramming.
TS6.8.Mb3 1983 6584'04 82-16035
ISBN 0-442-25415-6
PREFACE

The year 1983 marks well over a decade since the publication of the 2nd edition
of this text, and a quarter century since the initial development of PERT and
CPM. During this period, the methodology has matured considerably and the
emphasis on the early "frills" of these techniques has all but disappeared. While
the concepts of PERT probabilities and the elegant timecost tradeoff of CPM
were useful and helped to create credibility andinterest in these techniques dur-
ing their development and early application, emphasis has shifted towards prob-
lems deemed to be of more importance to practitioners. We have tended to
follow this movement in this third edition of the text. While we have not elimi-
nated the socalled "frills," we have reduced the space devoted to them some-
what, and have added new material that may improve their applicability.
Since the major benefit of network based methods of project planning and
control is the development of the project network itself, the most important
change in this thud edition is the emphasis on node and precedence diagrams
as alternatives t o the arrow diagram representation of a project plan. While the
arrow diagram was the mainstay of PERTICPM and their immediate offsprings
in the 1960s, the 1970s saw a gradual shift towards node diagrams, and more
recently to precedence diagrams. As the reader will learn, this is a mixed bless-
ing. The enrichment of network logic afforded by precedence diagramming
comes only at the expense of more complex results in the scheduling time anal-
ysis computations. We have included in this edition a comprehensive treatment
of this problem so that the reader can intelligently assess its potential in specific
application areas.
Another topic of increasing importance in the application of networkmethods
is the scheduling of activities under constraints on resource availablities. This
chapter of the text has been considerably expanded, and updated t o reflect the
current methods being used in this area. Finally, the use of computers in this
field has tended to follow the continued improvement and cost reduction in
computer hardware. The chapter on computer processing has also been ex-
panded and updated.
The basic organization of the chapters remains essentially the same as previous
editions, with Part I basic and Part I1 advanced subjects. The 6 chapters of Part
1 comprise a complete course in the fundamentals of the planning and scheduling
features of critical path methods, including cost control methods and practical
applications. One consistent set of terms and symbols is used throughout Part I.
Thus, industrial and commercial users of the methods may study only Part I in
preparation for most practical applications. A good two- or three-day training
course for industrial personnel can be based on Part I, with selected portions
from Part 11 as appropriate to the needs of the group.
The material in Part I1 is suitable for more advanced users and college level
courses in departments of management science and business administration, in-
dustrial and systems engineering, and civil engineering. While certain portions
of the text require some statistics and linear programming background, the
prerequisites are generally satisfied by upper level undergraduate students of
business or engineering.
Of interest to college instructors will be the added exercises at the end of
most chapters, and the inclusion of many problem solutions at the end of the
book.

Joseph J. Moder
Coral Gables, Florida
Cecil R. Phillips
Atlanta, Georgia
Edward W. Davis
Charlottesville, Virginia
NOMENCLATURE

A F = activity (free) float or slack.

a = the "optimistic" performance time estimate used in PERT-the time
which would be bettered only one time in twenty, i.e., the fifth per-
centile (where specifically noted, it will also be used to denote the
zero percentile used in conventional PERT).
b = t h e "pessimistic" performance time estimate used in PERT-the time
which would be exceeded only one time in twenty, i.e., the ninety-
fifth percentile (where specifically noted, it will also be used to de-
note the 100 percentile used in conventional PERT).
C=cost slope for an activity used in time-cost trade-off procedures.
Cd= direct costs associated with the performance of an activity in timed,
the "crash" performance time.
CD = direct costs associated with the performance of an activity in time D ,
the "normal" performance time.
d = "crash" activity performance time-the minimum time in which the
activity can be performed.
D = "normal" activity performance time-the one which minimizes the ac-
tivity direct costs; also used to denote the mean activity performance
time based on a single time estimate.
E = earliest (expected) event occurrence time.
ES = earliest (expected) activity start time.
EF = earliest (expected) activity finish time.
viii Nomenclature

F = total path float or slack.

FF= the leadllag time associated with a finish-to-finish constraint used in
precedence diagramming.
FS= the leadllag time associated with a finish-to-start constraint used in
precedence diagramming.
L =latest allowable cvent occurrence time.
LS = latest (allowable) activity start time.
LF= latest (allowable) activity finish time.
m = the "most likely" performance time estimate used inPERT-the modal
value of the performance time distribution.
SF= the leadllag time associated with a start-to-finish constraint used in
precedence diagramming.
SS = the leadllag time associated with a start-to-start constraint used in
precedence diagramming.
f = actual activity performance time, determined after the activity has ac-
tually been completed.
t, = mean activity performance time based on the three (PERT) time esti-
mates, a, m,and b.

the event has actually occurred.I

T = actual occurrence time of a speci IC network event, determined after

Td = total (expected) project duration time achieved by using "crash" ac-

tivity performance times on all critical path activities.
TD = total (expected) project duration time achieved by using all "normal"
activity performance times.
Ts= scheduled event occurrence time.
V, = the estimated variance of the actual performance time, t , based on the
PERT formula [(b - a)/3.2] '.
(v,)'I2 =square root of V,, called the standard deviation of the actual activity
performance time, t.
VT = t h e estimated variance of the actual occurrence time, T,of a specific
network event.
(vT)'I2 = square root of VT, called the standard deviation of the actual event oc-
currence time, T.
Z = standard normal deviate, equal to the difference between a random
variable, such as T, and its expected or scheduled time, such as Ts,
divided by the standard deviation of the random variable, such as
(~i")'" .
CONTENTS

Preface V

Nomenclature vii

PART I BASIC TOPICS

1 Introduction

Development of the Networkplan Concept

The Systems Approach to Project Management
Historical Development of Network-Based Project
Management Methodology
Summary of Network-Based Project Management Methodology
Uses of Critical Path Methods
Advantages of Critical Path Methods
References
Exercises

2 Development of the Network

Preparation for Networking

Basic Terms
Network Rules
 Emphasis on Logic
Interpretation of Rules
Common Pitfalls
Satisfying Computer Rules
Using Dummies Efficiently
Activity Descriptions
Sample Applications
Natural and Resource Dependencies
Time-Scaled Networks
Activity-On-Node Schemes
Basic Node Scheme
Precedence Diagramming
Summary
References
Exercises

3 Time Estimates and Level of Detail

Time Estimation in the Network Context

Estimation Methods
Activity Redefinition
Network Condensation
The Level of Detail
Arrow Versus Precedence Method
Summary
References
Exercises

4 Basic Scheduling Computations

Computation Nomenclature-Arrow Diagrams

Forward Pass Computations
Backward Pass Computations
Definition and Interpretation of Float (Slack)
Critical Path ~dentiication
Other Types of Float
Use of Special Symbols in Scheduling Computations
Illustrative Network Employing Special Symbols
Critical Path From Forward Pass Only
Variations of the Basic Scheduling Computations
 Illustrative Network with Multiple Initial and
Terminal Events
Networks with Scheduled or Directed Times on
Intermediate Events
Network Time-Status Updating Procedure
Activity-On-Node Scheduling Computations
Precedence Diagramming
Precedence Diagramming Computational Procedures
Discussion of Computations with Splitting Allowed
Summary of Basic Scheduling Computations
References
Exercises
Appendix 4-1 Algorithmic Formulation of Basic
Scheduling Computations
Topological Ordering of Project Activities
Computer Programming
Appendix 4-2 Algorithm for Basic Scheduling Computations
for Precedence Diagramming with Splitting Allowed

5 Project Cost Control

Basic Problems in Network-Based Cost Control

History of PERTICOST
Development of C/SCSC
Monitoring Project Costs: Value of Work Performed
An Example Operating System: Metier's ARTEMIS
Summary
References
Exercises

6 Comments on Practical Applications

Preparing Project Proposals

Contractual Requirements
Implementation
Project Control
Construction Industry Practices
Human and Organizational Issues
Coordinating Interdepartmental Activities
References
 PART II ADVANCED TOPICS

7 Resource Constraints in Project Scheduling

How Limited Resources Affect Schedule Slack

Multiproject Scheduling
Resource Loading
Overview of Basic Scheduling Procedures
Use of Heuristics in Scheduling
Parallel vs. Serial Methods
Resource Leveling
Fixed Resource Limits Scheduling
Optimization Procedures
Network Characteristics: Use in Resource-Constrained
Scheduling
Summary
References
Exercises
Appendix 7-1 Application of Branch and Bound Methods
to Resource Constrained Project Scheduling

8 Time-Cost Tradeoff Procedures

The Critical Path Method (CPM) of Time-Cost Trade-offs

Activity Time-Cost Trade-off Inputs for the CPM Procedure
Hand Computational Procedure
Economic Implications of Project Time-Cost Tradeoff
Summary
References
Exercises
Appendix 8-1 Applications of Mathematical Programming
to the Development of Project Cost Curves
Linear Programming Formulation of Basic Scheduling
Computations
Linear Programming Formulation of the CPM Time-Cost
Trade-Off Problem and Extensions

9 The Pert Statistical Approach

Overview of PERT
Basic Probability Theory
PERT System of Three Time Estimates
 Contents xiii

Probability of Meeting a Scheduled Date

Computation of Cumulative Probability vs. Project
Duration
Illustrative Example Using Percentile Estimates
Monitoring Activity Time Estimates and Performance
Summary of Hand Probability Computations
Investigation of the Merge Event Bias Problem
Ang's PNET Algorithm
Simulation Sample Size
Summary
References
Exercises
Appendix 9-1A and B The Cumulative Normal Distribution
Function
Appendix 9-2 Use of Historical Data in Estimating
a, m, and b

10 Other Networking Schemes

Development of GERT
Gert Nomenclature
An Application of Q-GERT
Results of Computer Simulation
Summary Discussion of GERT
References
Exercises

11 Computer Processing

Advantages of Computers
When a Computer is Needed
Computer Software
Input for Initial Computation
Output for Initial Computation
Input for Updating
Output for Updated Schedule
Precedence Method
How to Find Computer Services
Comparative Features
Trends in Software
Representative Software
xiv Contents

Summary
Software Vendor Addresses
References
Exercises

Solutions to Exercises
PROJECT MANAGEMENT
with CPM, PERT and
Precedence Diagramming
I
BASIC TOPICS
INTRODUCTION

Management is a process concerned with the achievement of goals or obiectives.

the
manager plans, organizes, staffs, directs, and controls to achieve an objective
with constraints on t i e , cost, and performance of the end product. This text
will deal primarily with these planning and control functions, and only peripher-
ally with organization matters. Planning is the p r o o f n g for the com-
mitment of resources in the most effective fashion. Controlling is the process of
making events conform to schedules by coordinating the action of all parts of
the organization according to the plan established for attaining the objective. It
can also be said that project management is a blend of art and science: the art of
getting things done through and with people in formally organized groups; and
the science of handling large amounts of data to plan and control so that project
duration and cost are balanced, and excessive and disruptive demands on scarce
resources are avoided. This text will deal primarily with the science of project
management, with occasional excursions into the art when it has a direct rela-
tionship with the science.
It is appropriate at this point to elaborate on the term project. A project is a
set of tasks or activities related to the achievement of some olanned obi=
normally where the objective is unique or non-repetitive. Thus, a project is usu-
ally distinguished from repetitive or continuous production processes by the
characteristic of uniqueness, or the "one-shot" nature of the objective. How-
ever, the term project will be interpreted quite broadly in this text in order to
4 1 1 Basic Topics

encompass the many possible applications of critical path methods, or network

planning techniques, that have come about since the development of PERT and
CPM in the late fifties. Projects may involve routine procedures, such as the
monthly closing of accounting books. In this case, network planning techniques
are useful for detailed analysis and optimization of the operating plan. Usually,
however, these techniques are applied to one-time efforts. Although similar
work may have been done previously, it is not being repeated in the identical
manner on a production basis. Consequently, in order to accomplish the project
goal or objective, the manager must plan and schedule largely on the basis of his
experience with similar projects, applying his judgment to the particular condi-
tions of the project at hand. During the course of the project, he must continu-
ally replan and reschedule because of unexpected progress, delays, or technical
conditions.
Until the advent of critical path methods, there was no generally accepted for-
mal procedure to aid in the management of projects. Each manager had his own
scheme, which often involved limited use of bar charts-a useful tool in produc-
tion management but inadequate for the complex interrelationships associated
with contemporary project management. The development of network based
planning methods in the late fifties provided the basis for a more formal and gen-
eral approach toward a discipline of project management. Critical path methods
involve both a graphical portrayal of the interrelationships among the elements
of a project, and an arithmetic procedure which identifies the relative impor-
tance of each element in the over-all schedule. Since their development, critical
path methods have been applied with notable success to research and develop-
ment programs, all types of construction work, equipment and plant mainte-
nance and installation, introduction of new products or services or changeovers
to new models, development of major transportation and energy related systems,
strategic long-term planning, management information systems developments,
production planning, emergency planning, and even the production of motion
pictures, conduct of political campaigns, and complex surgery. According to our
definition, all of these activities are classed as projects.
In all of these projects, management is concerned with developing a "good"
(or at least a workable) plan of the activities that make up the project, including
a specification of their interrelationships. AIso, management is interested in
scheduling these activities in an acceptable time span, considering the manpower
and other resources required to carry out the program as it progresses in time.
Management is also concerned with monitoring the expenditure of time and
money in carrying out the scheduled program, as well as the resulting "product"
quality or performance, for cases where achievement of the project objective(s)
can be measured on a continuous or ordinal scale(s). For the most part, critical
path methods have concentrated on the time parameter and to a somewhat lesser
extent on the cost parameter. The performance parameter is a much more diffi-
cult and varied problem. It will not be treated explicitly in this text.
 Introduction 5

DEVELOPMENT OF THE NETWORK PLAN CONCEPT

The network diagram is essentially an outgrowth of the bar chart which was de-
veloped by Gantt in the context of a World War 1 military requirement. The bar
chart, which is primarily designed to control the time element of a program, is
depicted in Figure I-la. Here, the bar chart lists the major activities comprising
a hypothetical project, their scheduled start and finish times, and their current
status. The steps followed in preparing a bar chart are as follows:

1. Analyze the project and specify the basic approach to be used.

2. Break the project down into a reasonable number of activities t o be
scheduled.
3. Estimate the time required to perform each activity.
4. Place the activities in sequence of time, taking into account the require-
ments that certain activities must be performed sequentially while others
can be performed simultaneously.
5. If a completion date is specified, the diagram is adjusted until this con-
straint is satisfied.

The primary advantage of the bar chart is that the plan, schedule, and progress
of the project can all be portrayed graphically together. Figure I -l a shows the
five-activity plan and 15-week schedule, and current status (end of third week)
indicates, for example, that activity B is slightly behind schedule. In spite of this
important advantage, bar charts have not been too successful onone-time-through
projects with a high engineering content, or projects of large scope. The reasons
for this include the following:

1. Planning and scheduling are considered simultaneously.

2. The simplicity of the bar-chart precludes showing sufficient detail to en-
able timely detection of schedule slippages on activities with relatively long
duration times.
3. The bar-chart does not show explicitly the dependency relationships
among the activities. Hence, it is very difficult to impute the effects on
project completion of progress delays in individual activities.
4. The barchart is essentially a manual-graphical procedure. It is awkward to
set up and maintain for large projects, and it has a tendency to quickly be-
come outdated and lose its usefulness.

With the above disadvantages in mind, along with certain events of the mid-
fifties, the stage was set for the development of a network-based project manage-
ment methodology. Some of the notable events were the emergence of general
systems theory and the second generation of large digital computers, and the
initiation of very large and technically demanding programs in weapons systems,
power generation systems, etc. Because of the enormous size of many of these
6 I I Basic Topics

Denotes the status;eg. activity A is ahead of

schedule at current timeend of week three (,I ciantt bar chart

I V denotes current status

Activities I
I
A I
V I
B. I
C I - Scheduled
proiect
D I
complerion
E I time
~imelnow
Time in
l I I l I I ~ ~ I (
0 / 3 4 5 6 7 8 9 10 11 12 13 14 15

(b) Proiect network

Time estimares

(c) Timescaled network

I
D-2 ---A I

0
i
1 2
l
3
I
4
1
5 6
1
7
l
8
i
9
1 ~ 1 ~1
10
1 1 1
11 12 13 14 15
- Time in
weeks
Figure 1-1 Comparison of G a n n chart, project network, and time-scaled network.
 Introduction 7

programs that contain thousands of significant activities, taking place in widely

dispersed locations, something like network-based methodology literally had to
emerge.
The heart of network-based planning methods is a graphical portrayal of the
plan for carrying out the program. Such a graph, called a network, shows the de-
pendency relationship among the project activities using the simple logic that all
activities preceding a given activity must be completed before thegiven activity
may begin. For example, in Figure I -l b , activities A-3 and E-l must both be
completed before E-2 may begin. Activities without predecessors (A-1, B-I, and
D-1) may begin anytime after the project starts, and when all of the activities
without successors (0-3 and E-2) are completed, the project is complete.
In the project network shown in Figure 1-lb,the lines denote activities which
usually require time, manpower, and facilities to complete. Each activity origi-
nates and terminates in a unique pair of nodes called events; time flows from the
tail to the head of each arrow. Events denote a point in time; their occurrence
signifies the completion of all activities terminating in the event in question. For
example, the occurrence of event 7 signals the completion of activities 3-7
(3 -* 7) and 6-7. Dashed line arrows, called dummies, show precedence relation-
ships only; they usually require no time, manpower, etc., to perform. Such a
relationship is shown by activity 7-5. This networking scheme is widely used
today and is called activity-on-arrows, or merely an arrow diagram. Another im-
portant way of networking is to reverse the role of the arrow and the node; the
result is called activity-on-nodes, or a node diagram. Detailed treatment of both
of these schemes of networking will be given in Part I of this text.
Figure I - l b shows the network plan for carrying out thin hypothetical project.
It should be noted that planning is separate from scheduling. The latter is ini-
tiated by adding estimates of activity duration times to Figure I-lb, and deter-
mining the impact of these times on overall project duration. One method of
showing this scheduling step is given in Figure 1-lc, where the project plan of
Figure I-lb is drawn t o scale on a time base. It is not too unlike the original
Gantt bar chart. It clearly illustrates, however, several major differences in the
traditional bar chart and time-scaled network diagrams. First, the network
shows greater detail. For example, activity D is broken down into activitiesD-1,
0-2, and 0-3. The second and more important difference is that the interdepen-
dency of the activities is clearly shown. For example, activity E can start as
soon as activity B is finished; however, the last portion of activity E, denoted by
E-2, cannot begin until activities E-1,A-3,C, and 0 - 2 are all completed.
In Figure I-lc, activities A-3,D-2, and E-2 have the last portion of their ar-
rows dashed. Based on the estimated activity duration times, these dashed lines
denote that these activities could be completed prior to the occurrence time of
their succeeding events. For this reason, the paths along which these activities
lie are referred to as slack paths, or paths withfloat time; that is, these paths re-
8 I I Basic Topics

quire less time to perform than the time allowed for them. For example, the
estimated time t o perform activities D-1 and 0 - 2 is 2 + 3 = 5 weeks, as indicated
on the time scale. However, the time interval between the occurrence of the ini-
tial and terminal events of this path, i.e., events 0 and 7, is 7 weeks. Thus, we
say that this path has two weeks of "slack" or "float." Now activities 17-1, C,
and 0 - 3 have no dashed portions; the sum of their expected performance times,
15, is the same as the time interval between the occurrence of events 0 and 8.
For this reason, this path is referred to as the criticalpath; it is the longest path
through the network.
From Figure 1-1c it is easy to see how the third of the above listed disadvan-
tages of the bar chart can be overcome. The impact of delays in completion of
any activity can be easily ascertained. For example, any delays in activities B-1,
C, and 0 - 3 will result in a corresponding delay in project completion, since they
have no slack. However, delays in activities D-1 plus 0 - 2 can be up to 2 weeks
before any delay in project completion would result. These concepts are treated
in detail in Chapter 4.
Although the network may be drawn t o a time scale, as shown in Figure 1-lc,
the nature of the network concept precludes this luxury in most applications, at
least in the initial planning stage. Thus, the length of the arrow is unimportant.
You can "slide" activities back and forth on a bar chart with ease, because the
dependency relations are not shown explicitly. However, if a network is drawn
on a time scale, a change in the schedule of one activity will usually displace a
large number of activities following it, and, hence, may require a considerable
amount of redrawing each time the network is revised or updated. There are
practical applications for time-scaled networks especially where the network is
closely related to production planning and where scheduling to avoid overload-
ing of labor or facilities is a prime consideration. In these cases the time-scaling
effort is worthwhile, for it clearly illustrates conflicting requirements.

THE SYSTEMS APPROACH TO PROJECT MANAGEMENT

Cleland and King1 * point out the strong influence of general systems theory, as
it had evolved in the mid-fifties, on the development of network-based project
management methodology. The systems concept instills the desire to achieve
overall effectiveness of the organization, in an environment which invariably in-
volves conflicting organuationalobjectives. A systems-oriented manager realizes
that he can achieve the overall goals of the organization only by viewing the en-
tire system. He must seek to understand and measure departmental interrela-
tionships, and to integrate them in a fashion which enables the organization to
efficiently pursue its goals. To accomplish this, increasing use is required of ob-
jective scientific analysis in solving decision problems. These methods rely on

These numbers refer to References given at the end of each chapter.

 Introduction 9

models-formal abstractions of real world systems-to predict the outcomes of

various available alternatives in complex decision problems. In effect, the sys-
tems approach may be viewed here as a logically consistent method of reducing
(by network based methods) a large part of a complex problem t o a simple out-
put. This output can be used by a decision maker, in conjunction with other
considerations, to arrive at a best decision. It permits him to put aside those
things which are best handled by systems analysis, and to focus on those aspects
of the problem which are most deserving of his individual attention. In this way,
the manager is able to get the "big picture" in its proper perspective, rather than
requiring him to devote attention to a myriad of minor, seemingly unrelated as-
pects of the total system.
Systems concepts have also brought about changes in the way plans are exe-
cuted. This change is primarily in organizational structures and the emergence
of the project manager, who must deal with traditional organizational philoso-
phy in a very delicate way. Traditional philosophy is based on a vertical flow of
authority and responsibility relationships, which emphasizes only parts and seg-
ments of the organization. It does not place sufficient importance on the inter-
relationships and integration of activities involved in the total array. In dealing
with these problems, the project manager, whose scope of interest and influence
cuts across these traditional vertical lines of authority, has been placed in a (for-
mal or informal) matrix form of organization. These concepts will be treated in
Chapter 6 of this text.
The strong influence of general systems theory on the evolution of the over-
all network-based methodology of project management can also be seen by an
analogy to systems control theory. The latter is illustrated in Figure 1-2a, where
it is applied to controlling the temperature of a room to a prescribed time-
temperature proffie, noted as INPUT. The control problem is created by ran-
dom OUTSIDE heat load DISTURBANCES impinging on the room to be con-
trolled, causing deviations from the desired INPUT. These deviations are detected
by a thermometer which monitors the state of the system, and transmits this in-
formation to the thermostat along the negative feedback loop. The thermostat
sends deviations between the INPUT and the OUTPUT to the controller, which
in turn drives the furnace in a manner t o eliminate the undesired temperature
deviations.
The analogy for network-based project management methodology is shown in
Figure 1-2b. The control system INPUT is the network-based project plan, in-
cluding a time and cumulative cost schedule of the project activities. The con-
trol problem here is created by random OUTSIDE DISTURBANCES to material
procurements, adrninistrativeltechnical approvals, personnel and technical prob-
lems, etc. The effects of these disturbances on the project are detected by a
periodic monitoring of activity time and cost status in the form of field reports.
The latter are used to update the project network t o determine the current time
and cost status of the project. Deviations from the INPUT plan are the basis for
10 I / Basic Topics

Random
outside heat
load distutbances

Furnace
AIC

Temperature
Negative feedback loop

(a) Cantrol theory applied to room temperature control

-- --

Random outside
disturbances:
procurement delays,
personnel problems,
technical problem.
weather effects

( b ) Network-based project management control methodology

Figure 1-2 Control t h e o r y analogy t o network-based p r o j e c t management methodology.

action by the project manager who initiates control decision actions, as needed,
to bring the project back into line. This is the basic methodology to be devel-
oped in this text.

HISTORICAL DEVELOPMENT OF NETWORK-BASED

PROJECT MANAGEMENT METHODOLOGY

As mentioned above, the initiation of large development programs in the 1950s,

along with the simultaneous development of the digital computer and general
systems theory, set the stage for the development of network-based project man-
agement methodology. Actually, it was a rediscovery of procedures that can at
least be traced back a half century t o a Polish scientist named Karol Adamiecki
(1931).' He developed and published a methodology in a form that he called
the Harmonygraph. This is a 90-degree rotated bar chart type graph with a verti-
cal time scale, and a column (strip) for each activity in the project. It is illus-
 Introduction 11

trated in Figure 1-3, using the same network adopted in Figure 1 - 1 . Note that
the activity strips are ordered so that the predecessors of any given activity will
always be found to its left. The activity strips each contain a movable tab whose
length is proportional to the estimated duration of the activity, and whose loca-
tion along the time scale denotes the scheduled start and finish time for each ac-
tivity. Thus, starting at the first column on the left in Figure 1-3 and working to
the right, it is easy to schedule each activity so that all time precedence con-
straints are satisfied. This is carried out by noting that constraining completion
times of all predecessor activities will have already been determined, and will ap-
pear nearby t o the left. For example, activity E-2, in Figure 1-3, has an earliest
allowable start time of 11, since the earliest finish times of its predecessorsA-3,
E-1, and Dummy, are 9, 11 and 7, respectively. The largest, or latest, of these
times is 1 I, which becomes the earliest allowable start time for activity E-2.
The Harmonygraph is equivalent to an arrow diagram of the project plan, and
hence it could alleviate all of the shortcomings of the bar chart listed above. By
incorporating sliding tabs on each activity strip, it even overcomes the difficulty
of keeping a bar chart up to date. These tabs facilitate the simulation of alterna-
tive schedules and network updating to reflect current status deviations from
original schedules.

*The first column or strip represents ACTIVITY A-1, where (4) indicates the estimated time
to perform this activity. The dash in the FROM row indicates that activity A-1 has no pre-
decessor activities, and the A-2 in the T O row indicates that it is a successor to A-1.
Figure 1-3 Harmonygraph drawing of the project depicted in Figure 1-1, with all activities
placed at their earliest start times.
12 1 / Basic Topics

One can only conjecture why this elegant solution to the problem was ig-
nored, and had to be reinvented some years later. First, the author, while de-
veloping a fine methodology, did very little to sell its merits with impressive ap-
plications. But the most important reason was undoubtedly the timing of the
development; in 1930 it was ahead of its time by a quarter of a century.
After these 25 years passed, an explosion occurred in interest in the problem
of scheduling large projects. Three developments in the U.S. and Great Britain
will be described below. Although they were independent, concurrent studies,
they all based their respective methodologies on a network representation of the
project plan, such as that shown in Figure 1-lb. They all essentially rediscovered
the Harmonygraph in the systems approach age of the computer, and added one
essential feature: They broke away from a strict graphical procedure by adding a
tabular and arithmetic approach to the scheduling process.
The unpublished British development is discussed by L ~ c k ~ e He r.~describes
how the Operational Research Section of the Central Electricity Generating
Board investigated the problem involved with the overhaul of generating plants.
By 1957, they had devised a technique which consisted essentially of identifying
the "longest irreducible sequence of events." This term was later shortened to
"major sequence," and corresponds to what we have referred to above as the
critical path. Successive experimental applications of this methodology in the
period 1958-60 resulted in very impressive project duration reductions of 42%
and then 32%of the previous average times.
While this study was being conducted in Great Britain, two corresponding de-
velopments were taking place in the U.S. They are called PERT and CPM, and
the historical highlights of each will be given below.
The development of PERT began when the Navy was faced with the challenge
of producing the Polaris missile system in record time in 1958. Several studies4'
indicated that there was a great deal to be desired with regard to the time and
cost performance of such projects conducted during the 1950's. These studies
of major military development contracts indicated that actual costs were, on the
average, two t o three times the earliest estimated costs, and the project durations
averaged 40 to 50 percent greater than the earliest estimates. Similar studies of
commercial projects indicated average cost and time overruns were 70 and 4 0
percent, respectively. While many people feel that original estimates must be
optimistic in order to obtain contracts, a more important reason for these fail-
ures was the lack of adequate project management planning and control tech-
niques for large complex projects.
It was recognized that something better was needed in the form of an inte-
grated planning and control system for the Polaris Weapons System program. To
face this challenge, a research team was assembled consisting of representatives
of Lockheed Aircraft Corporation (prime contractor of Polaris), the Navy Spe-
cial Projects Office, and the consulting firm of Booz, Allen and Hamilton. This
research project was designated as PERT, or Program Evaluation Research Task.
 Introduction 13

By the time of the first internal project report, PERT had become Project Eval-
uation and Review T e ~ h n i ~ u eThis. ~ research team evolved the PERT system
from a consideration of techniques such as Line-of-Balance,' Gantt charts, and
milestone reporting systems.
Time was of the essence in the Polaris program, so the research team concen-
trated on planning and controlling this element of the program. A major accom-
plishment of the PERT statistical procedure is the utilization of probability
theory for managerial decision making. Scheduling systems have, traditionally,
been based upon the idea of a f ~ e dtime for each task. In the PERT system
three time estimates are obtained for each activity-an optimistic time, a most
likely time, and a pessimistic time. This range of times provides a measure of the
uncertainty associated with the actual time required to perform the activity
sometime in the future. With the PERT procedure, it is possible, on the basis of
these estimates, to derive the probabilities that a project will be completed on or
before a specified schedule date. The misleading notion of a definite time for
the completion of a project, or subproject, can be replaced by statements of the
possible range of times and the probabilities associated with each. The result is
a meaningful and potentially useful management tool. By adding to this infor-
mation an appraisal of the consequences of not meeting a scheduled date and the
cost of expediting a project in various ways, management can better plan at the
outset of a project.
PERT also emphasizes the control phase of project management by various
forms of periodic project status reports. The work of the original PERT research
team has been extended into the areas of planning and controlling c o s t s , 8 ~and
~
to a lesser degree, into the areas of the performance or quality of the product.1°
CPM (Critical Path Method) grew out of a joint effort conducted in the period
Dec. 1956-Feb. 1959 by the duPont Company and Remingt~nRandUnivac."~'~
The objective of the CPM research team was to determine how best to reduce
the time required to perform routine plant overhaul, maintenance, and constmc-
tion work. In essence, they were interested in determining the optimum trade-
off of time (project duration) and total project cost. This objective amounts to
the determination of the duration of a project which minimizes the sum of the
direct and indirect costs, where, for example, direct costs include labor and ma-
terials, while indirect costs include the usual items, such as supervision, as well
as "cost" of production time lost due to plant downtime.
The activities comprising this type of project are characteristically subject to a
relatively small amount of variation compared to the activities of the Polaris pro-
gram. Hence, unlike PERT, CPM treats activity performance times in a deter-
ministic manner and has as its main feature the ability to arrive at a project
schedule which minimizes total project costs.
The pioneering PERT and CPM groups did not know of each other's existence
until early 1959, when the momentum of each effort was too great to influence
the other. What has finally emerged from them is essentially a methodology
14 I 1 Basic Topics

similar to the Harmonygraph, with the addition of tabular computer outputs

that give the start and finish times and slack (float) for each activity, and the
need to sort this output in a variety of ways for different people to use. The
sophisticated time-cost trade-off optimization algorithm of CPM, and the prob-
ability of meeting a schedule feature of PERT really play a very minor role jn
the applications today. However, it is felt by Kelley,13 one of the pioneers of
CPM, that we might not have CPM or PERT today if it were not for these sophis-
ticated frills. They added that "something extra" to heighten interest and moti-
vation and, more importantly, to enhance credibility. Although it may only be a
spurious correlation, it is interesting t o note that the Harmonygraph lacked such
a frill, and it never made it in the world of applications.
This brief historical treatment of network based project management meth-
odology is by no means complete. For example, the work of John Fondahl,14
commencing in 1958 is certainly noteworthy. It dealt with a node-diagram ap-
proach to the basic CPM problem. Also, about this time, the development of the
"method-of-potentials" took place in France. It was based on a network logic
which constrained the start of one activity to lag a specified amount of time
after the start of its predecessor activity. In a sense, this was a forerunner to the
development of "precedence diagramming" which took place in the U.S.in the
1960s. Other studies of note will be cited in subsequent chapters.

SUMMARY OF NETWORK-BASED PROJECT

MANAGEMENT METHODOLOGY

Network-based project management methodology is a dynamic planning and

control procedure, as was illustrated in Figure 1-2 above. This concept is shown
in embellished form in Figure 1-4, which embodies the following steps, each of
which will be described below:

STEP 1 . Project Planning

STEP 2. Time and Resource Estimation
STEP 3. Basic Scheduling
STEP 4. Time-cost Trade-offs
STEP 5. Resource Allocation
STEP 6. Project Control

Step 1. Project Planning

The activities making up the project are defmed, and their technological depen-
dencies upon one another are shown explicitly in the form of a network dia-
gram. This step is shown in box (1) of Figure 1-4, and is the subject of Chapter
2, Developing the Network. Three alternative methods of networking will be
 Introduction 15

I
-
-

Initial (11 Network diagram (2) Individual activity

showing plan of work
duration and resource
and technological
specifications estimates
constraints

i
t (3) Basic critical PLANNING
Modifications t o time, +
path calculations PHASE
cost, performance, or
resource constraints

t
-
+
(7)Management
replanning and overall NY 14) Resolution
Of
and time
resource allocation constraints
No Yes

IS current project (51 Acceptable plan

plan satisfactory? for implementation

t CONTROL
(6) Periodic status PHASE
,
reports: elapsed time.
measured progress,
resources expended.
money expended, etc.
-
Fiiure 1-4 Dynamic network-based planning and control procedure.

taken up, called m w ,node, and precedence diagramming. Thii is the most im-
portant step in the entire PERTlCPM procedure. The disciplined approach of
expressing a plan for carrying out a project in the form of a network accounts
for a large portion of the benefits to be derived from PERTICPM. The develop-
ment of the network is, in a sense, the simulation of alternative ways of carrying
out the project. Experience has shown that it is preferable to make planning er-
rors on paper, rather than in "bricks and mortar." It should also be added that
if useful results are not obtained from these methods, it is usually because of
inadequately prepared networks.

Step 2. Time and Resource Estimation

Estimates of the time required to perform each of the network activities are
made; these estimates are based upon assumed manpower and equipment require-
ments and availabiiity, and other assum~tionsthat mav have been made in dan-
ning the project in step 1. This step isshown in boxi2) of Figure 1-4. ~;ngle-
time estimation is taken up in Chapter 3. The three-time estimation method
associated with PERT is treated in Chapter 9.
16 l l Basic Topics

Step 3. Basic Scheduling

The basic scheduling computations give the earliest and latest allowable start and
finish t i e s for each activity, and as a byproduct, they identify the critical path
through the network, and indicate the amount of slack or float time associated
with the noncritical paths. This step, shown in box (3) of Figure 1-4, is taken
up in Chapter 4.

Step 4. Time-Cost Trade-offs

If the scheduled time to complete the project as determined in Step 3 is satisfac-

tory, the project planning and scheduling moves on to a consideration of re-
source constraints in Step 5. However, if one is interested in determining the
cost of reducing the project completion time, then time-cost trade-offs of activ-
ity performance times must be considered for those activities on the critical and
near critical paths. This step, shown in box (4) of Figure 1-4, is taken up in
Chapter 8.
The costs associated with a project can, for certain purposes, be classed as
either direct or indirect. The direct costs typically include the items of direct
labor and materials, or if the work is being performed by an "outside" company,
the direct costs are taken as the subcontract price. The indirect costs may in-
clude, in addition t o supervision and other customary overhead costs, items such
as the interest charges on the cumulative project investment, and penalty (or
bonus) costs for completing the project after (or before) a specified date. The
time-cost trade-off problem is directed to the task of determining a schedule of
project activities which considers explicitly the indirect as well as the direct
costs, and attempts to minimize their sum.

Step 5. Resource Allocation

The feasibaity of each schedule must be checked with respect to manpower and
equipment requirements, which may not have been explicitly considered in Step
3. This step, shown in box (4) of Figure 1-4, is taken up in Chapter 7. The rou-
tine scheduling computations of Step 3 determine the slack along each network
path. This indicates where certain activity schedules can be moved forward or
backwards in time without affecting the completion time of the project. This
movement can then be used to arrive at schedules which satisfy outside con-
straints placed on the quantity of resources available as a function of time.
Establishing complete feasibility of a specific schedule requires frequent rep-
etition of the basic scheduling computations, as shown by the recycle path from
box (4) to box (3). It may also require replanning and overall adjustment of re-
sources, as shown by the path from box (4) to box (7). Hence, establishing an
acceptable project plan for implementation may require the performance of a
number of cycles of Steps 3 and 4, and possibly Steps I and 2 as well.

Step 6. Project Control (Time and Cost)

When the network plan and schedule have been developed to a satisfactory ex-
tent, they are prepared in final form for use in the field. The project is con-
trolled by checking off progress against the schedule, as indicated in box (6), and
by assigning and scheduling manpower and equipment, and analyzing the effects
of delays. Whenever major changes are required in the schedule, as shown in box
(7), the network is revised accordingly and a new schedule is computed. The
subject of time control is taken up in Chapter 4, and cost control in Chapter 5.
In addition to its value as a means of planning a project t o optimize the time-
cost relationship, the critical path network provides a powerful vehicle for the
control of costs throughout the course of the project. Most cost accounting sys-
tems in industry are functionally-oriented, providing cost data by cost centers
within the company organization rather than by project. By the utilization of
the project network as a basis for project accounting, expenditures may be
coded to apply to the activities, or groups of activities within a project, thus en-
abling management to monitor the costs as well as the schedule progress of the
work.
Although the theory of network cost control is relatively simple, it is just be-
ginning to be employed as a practical supplement to basic critical path technol-
ogy, primarily because of the necessary involvement in established cost account-
ing procedures, and the fairly recent widespread availability of general computer
programs. Thus, each organization interested in network cost control has been
faced with the inconvenience and expense of developing new accounting proce-
dures and the adoption of appropriate computer programs. In the mid-60s sev-
eral large agencies of the U.S. Governments required the use of cost control
supplements to basic CPM and PERT requirements, and generalized computer
programs were developed. The U.S. Army Corp of Engineers still has this re-
quirement on complex projects. They have not, however, succeeded in making
cost control supplements as widely used as the basic PERTICPM procedures, but
they did provide some impetus for continued development of cost control sys-
tems. Also, as the installation of computers has expanded further in industry,
changes in accounting procedures have become more frequent. Indeed, this ex-
pansion is becoming a necessity in many functionally organized companies that
are experiencing problems of coordinating project activities because of rapid
expansion of the volume of their work. Much of this has been brought about by
the frequent occurrence of multibillion dollar projects contracted on a cost basis.
Such firms are currently expending a great deal of effort to develop network-
based management information systems to alleviate this problem.
 An introduction to the concepts and practical problems of network cost con-
trol is presented in Chapter 5 . A discussion of the available computer programs
for all types of network analysis is contained in Chapter 11.
The basic procedures incorporated in Steps 1 through 6 can be performed, at
least to some extent, by hand. Such methods will be presented in this text be-
cause they are useful in their own right, and also because they are an excellent
means of introducing the more complex procedures that require the use of com-
puters. It is particularly important that one be able to perform, by hand, the
basic critical path calculations indicated in box (3) of Figure 1-4, since this is the
first step in the evaluation of a proposed network plan for carrying out a project.
A very simple method of hand calculation will be presented in Chapter 4. Hand
methods for the resolution of relatively simple time and resource constraints will
also be presented; they will then lead into more complex procedures for which
computers are a necessity. Hand and computer methods of preparing periodic
status reports will also be presented. A description of available computer pro-
grams is given in Chapter 11.

USES OF CRITICAL PATH METHODS

Since the successful application of PERT in the Polaris program, and the initial
success of CPM in the chemical and construction industries, the use and further
development of critical path methods has grown at a rapid rate. The applica-
tions of these techniques now cover a wide spectrum of project
Research and development programs range from pure research, applied re-
search, and development to design and production engineering. While PERT is
most useful in the middle of this spectrum, variations of it have been used in the
production end of this spectrum. However, overall usage is not as widespread
now as in the 1960s. PERT is not particularly useful in pure research, and in
fact some say it should be avoided here because it may stifle ingenuity and imag-
ination, which are the keystones of success in pure research.
Maintenance and shutdown procedures, an area in which CPM was initially de-
veloped, continues t o be a most productive area of application of critical path
methods. Construction type projects continue to be the largest individual area
in which these methods are applied. It is extremely useful in this field of appli-
cation t o be able to evaluate alternate project plans and resource assumptions on
paper rather than in mortar and bricks.
More recent applications of critical path methods include the development
and marketing of new products of all types, including such examples as new
automobile models, food products, computer programs, Broadway plays, and
complex surgical operations.
In addition t o an increase in the variety of applications of critical path meth-
ods, they are being extended to answer questions of increasing sophistication.
The important problem of resource constraints has been successfully expanded
 lntroductlon 19

to include multiple resource types associated with multiple projects. Cost con-
trol, project bidding, and incentive contracting are also areas where significant
developments are taking place.

ADVANTAGES OF CRITICAL PATH METHODS

It is fitting to close this chapter with an enumeration of the advantages that one
might expect from the use of critical path methods in the planning and control-
ling of projects.
1 . Planning Critical path methods first require the establishment of project
objectives and specifications, and then provide a realistic and disciplined
basis for determining how t o attain these objectives, considering pertinent
time and resource constraints. It reduces the risk of overlooking tasks nec-
essary to complete a project, and also it provides a realistic way of carrying
out more long-range and detailed planning of projects, including their coor-
dination at all levels of management.
2 . Communication Critical path methods provide a clear, concise, and un-
ambiguous way of documenting and communicating project plans, sched-
ules, and time and cost performance.
3. Psychological Critical path methods, if properly developed and applied,
can encourage a team feeling. It is also very useful in establishing interim
schedule objectives that are most meaningful t o operating personnel, and
in the delineation of responsibilities to achieve these scheduled objectives.
4 . Control Critical path methods facilitate the application of the principle
of management by exception by identifying the most critical elements in
the plan, focusing management attention on the 10 to 20 per cent of the
project activities that are most constraining on the schedule. I t continu-
ally defines new schedules, and illustrates the effects of technical and pro-
cedural changes on the overall schedule.
5 . Training Critical path methods are useful in training new project mana-
gers, and in the indoctrination of other personnel that may be connected
with a project from time to time.

REFERENCES

1. Cleland, D. I. and W. R. King, Systems Analysis and Project Management,

McGraw Hill Book Co., Inc., 1968.
2. Adamiecki, Karol, "Harmonygraph," Przeglad Organizacji (Polish Journal
on Organizational Review), 1931.
3. Lockyer, K. G., Introduction to Critical Path Analysif, Pitman Pub. Co.,
3rd Edition, 1969, Ch. 1.
4. Marshall, A. W., and W. H. Meckling, "Predictability of the Costs, Time and
Success of Development," RAND Corp., Report P-1821, December, 1959.
 5. Peck, M. J. and F. M. Scherer, "The Weapons Acquisition Process: An Eco-
nomic Analysis," Division of Research, Graduate School of Business Admin-
istration, Harvard University, Cambridge, Mass., 1962.
6. Malcolm, D. G., J. H. Roseboom, C. E. Clark, and W. Pazar, "Applications
of a Technique for R and D Program Evaluation," (PERT) Operations Re-
search, Vol. 7, No. 5, 1959, pp. 646-669.
7. Turban, Efraim, "The Line of Balance-A Management by Exception Tool,"
The Journal o f Industrial Engineering, Vol. 19, No. 9 , September 1968, pp.
440-448.
8. The Office of the Secretary of Defense and the National Aeronautics and
Space Administration, DUD and NASA Guide, PERT Cost Systems Design,
U.S. Government Printing Office, Washington, D.C., June, 1962, Catalog
Number Dl. 6/2:P94.
9. Moder, J. J. and S. E. Elmaghraby, Handbook of Operations Research Mod-
els and Applications, Van Nosband Reinhold Co., New York, NY, 1978,
Chapter 10.
10. Malcolm, D. G., "Reliability Maturity Index (RM1)-An Extension of PERT
into Reliability Management," The Journal o f Industrial Engineering, Vol.
14, No. 1, January-February 1963, pp. 3-12.
11. Walker, M. R. and J. S. Sayer, "Project Planning and Scheduling," Report
6959, E. I. duPont de Nemours and Co., Wimington, Delaware, March
1959.
12. Kelley, J., "Critical Path Planning and Scheduling: Mathematical Basis,"
Operations Research, Vol. 9, No. 3, May-June 1961, pp. 296-321.
13. Kelley, J., CPM-In Conception or An Avowing Experience, Unpublished
talk.
14. Fondahl, J. W., "A Noncomputer Approach to the Critical Path Method for
the Construction Industry," Dept. of Civil Engineering, Stanford University,
Stanford, Calif., 1st Edition, 1961, 2nd Edition, 1962.
15. Berkwitt, G. J., "Management Rediscovers CPM," Dun's Review, May 197 1,
Dun & Bradstreet Pub. Corp.
16. Davis, E. W., "CPM Use in Top 400 Construction Firms," Journal o f the
Construction Division, ASCE, Vol. 100, No. CO1, Proc. Paper 10395, March
1974, pp. 39-49.

EXERCISES

1. Discuss various applications of critical path methods. For example, suppose

you are in charge of the preparation of a proposal for a large and involved
project, or the coming church social, or the preparation of a new college cur-
riculum, or the development of a new product and manufacturing facility.
Would critical path methods be of assistance in these undertakings? If so, in
what ways?
2. Can you think of any complex projects in which critical path methods would
not be of any particular value? Give examples and discuss why.
DEVELOPMENT
OF THE
NETWORK

The first step in utilizing critical path methods is the identification of all the ac-
tivities involved in the project and the graphical representation of these activities
in a flow chart or network. This step is usually called the "planning phase," be-
cause the identification of the project activities and their interconnections re-
quires a thorough analysis of the project, and many decisions are made regarding
the resources to be used and the sequence of the various elements of the project.
In one sense the network is only a graphical representation or model of a proj-
ect plan. The plan may have previously existed in some other fom-in the
minds of the project supervisors, in a narrative report, or in some form of bar
chart. In practice, however, the preparation of a network usually influences the
actual planning decisions and results in a plan that is more comprehensive, con-
tains more detail, and is often different from the original thoughts about how
the project should proceed. These changes derive from the discipline of the net-
working process, which requires a greater degree of analytical thinking about the
project than does a narrative, a bar chart, or other types of project descriptions.
Thus, the contruction of the network often becomes an aid to and an integral
part of project planning, rather than an after-the-fact graphical exercise. Indeed,
the planning phase has proven to be the most beneficial part of critical path ap-
plications. In developing a detailed and comprehensive project network, users
often make significant improvements over their original ideas; they do a better
22 I 1 Basic Topics

job of early coordination with suppliers, engineers, managers, subcontractors,

and all the other groups associated with the project; and they end up with a
documented plan that has strong psychological effects on the future manage-
ment of the project.
One psycholo~icaleffect of network preparation is that it demonstrates to
shpervisors and other key personnel that&man_agm2nt is vitally concerned
-
about the coordination and timeliness of all project a c t i v i t m t h a t a means
of more closely monitoring these factors has been drafted. Thus, an intangible
but highly significant factor-the initial motivation of the project team-can he
favorably influenced by the networking effort.
The planning phase is also the most timeconsuming and difficult part of most
critical path method applications. This is due primarily to the inherent analyti-
cal problems in any project planning effort. One may expect some difficulty in
using the network format at first, but it is soon realized that the network disci-
pline is more of an aid to thinking than it is a set of stringent rules for drawing a
chart.
Actually there are only about five rules commonly used in drawing networks,
and these provide most of the flexibility needed in describing project plans.
(More advanced networking schemes with even greater flexibility are described
in Chapter 10.) The accuracy and usefulness of a network is dependent mainly
upon intimate knowledge of the project itself, and upon the general qualities of
judgment and skill of the planning personnel. Skill in using the network tech-
niques can be quickly acquired in only one or two applications. The time re-
quired to learn to use networks drawn by someone else is even less.
This chapter is limited to those basic rules and procedures of network devel-
opment which are required t o prepare the first draft of a network. In Chapter 3
the addition of time estimates and the development of a final working draft are
considered, including the problem of obtaining the most useful level of network
detail.
There are several different graphical schemegused in drawing networks. This
chapter and tre rest omart I will be based upon the most common scheme prac:
ticed among industrial users of critical path methods. This is not necessarily the
test scheme, however. Two of the other schemes, the node scheme andgrece-
dence diagramming, have much t o recommend them, especially under certain
conditions. Readers who are not limited by contract, the computer programs
available, or other reasons to use a particular scheme are urged to study all of the
schemes, which are presented in this chapter and in Chapter 3. The reader
should then select the scheme best suited t o the circumstances. As mentioned
above, in the application of critical path methods the greatest effort and expense
are associated with the preparation of the network. Selection of the most useful
and economical networking scheme is, therefore, worthy of special attention.
 Develo~mentof the Network 23

PREPARATION FOR NETWORKING

Some experienced network users will say that all one needs t o begin networking
is a large piece of paper, several sharp pencils, and a large eraser. Actually, there
is a bit more to it than that. Several general questions need to be raised and
answered before detailed project planning should begin. Among these questions
are:

1. What are the project objectives?

2. Who will be charged with the various responsibilities for accomplishing the
project objectives?
3. What organization of resources is available or required?
4. What are the likely information requirements of the various levels of man-
agement to be involved in the project?

Of course, these questions are fundamental t o project management and should

not be passed over lightly. In some research and development projects, the de-
velopment of new products, and in other cases, a discussion of the basic objec-
tives of the project can reveal disagreements among the key persons involved.
Similarly, open discussions of responsibilities and resources can bring to light
erroneous assumptions or misunderstandings in these areas. Naturally, it is well
to resolve these matters before proceeding with networking (and certainly before
beginning the project).
An optional step in network preparation is the development of a list of work
elements of the project. Such a list can be useful in discussing responsibilities
and resources, as mentioned above, and it can serve as a reference for network-
ing. Although experienced networkers usually forego the listing of activities, be-
ginners with the technique will find that a list is helpful.

BASIC TERMS

Several of the most common terms in networking are defined and illustrated
below. Terms associated with scheduling computations are explained in later
chapters.

Definition:

synonymous with "activity" are "task" and 'Sob." In the arrow scheme of net-
 24 l l Basic Topics

Prepare a m r k
t
8 days

Figure 2-1

--------*
Figure 2-2a

Figure 2-26

working, activities are graphically represented by arrows, usually with descrip-

tions and time estimates written along the arrow (Figure 2-1).

Definition:

An arrow representing merely a dependency of one activity upon another is

called a dummy activity. A dummy carries a zero time estimate. It is also called
a "dependency arrow." Dummies are often represented by dashed-line arrows
-(Figure 2-2a) or solid arrows with zero time estimates (Figure 2-2b).

Definition:

The beginninp and ending points of activities are called events. Theoretically,
an event is an instantaneous point in time. Synonyms are ''node" and "connec-
.
tor." If an event represents the joint completion of more than one activity, it is
called a "merge" e m . If an event represents the joint initiation of more than
one activity, it is called a "burst" event. An event is often represented graphi-
cally by a numbered circle (Figure 2-3); although any geometric f ~ u r will
e serve
the purpose.

Event Burst event

Figure 2-3
 Development of the Network 25

Figure 2-4

Definition:

A network is a graphical representation of a project plan, showing the inter-

relationships of the various activities. Networks are also called "arrow diagrams"
(Figure 2-4). When the results of time estimates and computations have been
added t o a network, it may be used as a project schedule.

NETWORK RULES

The few rules of networking based on the arrow scheme may be classified as
those basic to all arrow networking systems, and as those imposed by the use of
computers or tabular methods of critical path computation.

Basic Rules of Network Logic

KULE I . Bcforc an activity may begin, all activities preceding it must be com-
pleted.
- . (Activities with n o ~rcdecessorsare self-actuating- whcn the
project begins.)
RULE 2. Arrows imply logical precedence only. Neither the length of the arrow
nor its "compass" direction on the drawing have any significance. (An
exception to this rule is discussed under "Time-scaled Networks"
below.)

Additional Rules Imposed By Some Computers or Tabular Methods

RULE 3. Event numbers must not be duplicated in a network.

RULE 4. Any two events may be directly connected by no more than one
activity.
26 1 I Basic Topics

R U L E 5. Networks may have only one initial event (with no predecessor) and
only one terminal event (with no successor).

Rules 4 and 5 are not required by all computer programs for network analysis,
as discussed in the appendix to Chapter 4. Hand computation on the network
does not require Rules 3 , 4 , or 5, but the network must not have loops.

EMPHASIS ON LOGIC,

At this point, it should be noted that the con&u&ion of a network should be

based on the logiial or technical dependencies among the activities. That is, the
activity "approve shop drawings" must be preceded by the>vity "prepare
shop drawings," because this is the logical and technically necessary sequence.
A common error in this regard is to introduce activities into the network on
the basis of a sense of time, or a "feel" for appropriate sequencing. For exam-
ple, in the maintenance of a pipeline the activity "deactivate lines" might be
placed after "procure pipe," because it is felt that that is the right time to deacti-
vate the lines. (Figure 2-5a.) Rather, the deactivation activity should be placed
in the network in the proper technological sequence, such as just before "remove
old pipe" (Figure 2-5b). Then in the scheduling process (to be covered in Chap-
ter 4) the best time to initiate the deactivation so as t o minimize the down time
on the pipelines can be determined.
Such emphasis on strict logic is one of the principles of networking intro-
duced by the originators of both CPM and PERT. It is a fundamental part of the
networking discipline that causes planners to think about their projects in a
thorough, analytical manner. In this process old methods of performing similar
projects may be questioned or disregarded, clearing the way for new and perhaps
better approaches.
Application of strict logic also tends to result in a unique network for a given
set of activities, that is, the network that represents the true technical dependen-
cies of the project. If a planner introduces his or her personal "feel" for how a

Procure Deactivate Remove old Install

lines new pipe

Figure 2-58

- Deactivate
Procure pipe

Remove old Install

lines pipe new pipe
u

Figure 2-5b
 Development of the Network 27

project should proceed, without strict attention to the logical dependencies,

than a subjective network is produced. Subjective networks may go unnoticed
for a while, but when the project gets underway and progress is reported, the
subjective relationships tend to be revealed in embarrassing ways. Activities be-
gin before the network says it is possible, others are delayed for activities that
are supposed to be independent, etc. The subjective network thus loses credibil-
ity with the users and may be abandoned early in the updating process.

INTERPRETATION OF RULES

Rules 1 and 2 may be interpreted by means of the portion of a network shown

in Figure 2-6a. According t o Rule 1, this diagram states that "before activity D
can begin, activities A , B , and C must be completed." Note that this is not in-
tended to imply that activities A , B, and Cmust be completed simultaneously.
Note also the definition of the events. Event '5 represents the "beginning of
activity A." Event 6, however, means "the completion of activities A , B , and C,
and the beginning of activity D." Because of the multiple meanings of events,
discussion of networks in terms of activities is favored over event-oriented terms.

COMMON PITFALLS

The most common network error involves Rule 1. As an illustration, consider

the diagram of activities A , B , C, and D shown in Figure 2-6a. Suppose that ac-
tivity D depended on the completion of B and C and on the completion of the
first half of A , completing the second half of A being independent of B, C, and
D. To diagram this situation correctly, we must divide activity A into two activi-
ties and introduce a dummy activity, as shown in Figure 2-6b. The dummy has
been used here to correct a problem of false dependency; that is, activity D was

Figure 2-6a
28 1 I Basic Topics

First half of A

Figure 2-66

only partially dependent on the activities preceding D. False dependencies rep-

resent the most subtle networking problems and must be guarded against con-
stantly, especially at merge and burst points. (The problem of false depen-
dencies is largely avoided in the node scheme of drawing networks, which is
discussed later.)
Another network condition that must be avoided is illustrated in Figure 2-7.
-
Activities J, F, and K form a 10% which g a n indication of faulty logic. The
definition of one or more of the dependency relationships is not valid. Activity
J cannot begin until C and K are completed. But K depends on F, which de-
pends on J. Thus J could never get started because it depends on itself. Loops,
which in practice may occur in a complex network through oversight, may be
remedied by redefining the dependencies t o relate them correctly.

SATISFYING COMPUTER RULES

Networking rules 3, 4, and 5 are related to the procedures for coding networks
for computer analysis. Rule 3 involves another subtle problem that all computer
programs have in understanding a network. Consider the diagram in Figure 2-8a.
 Develonrnent of the Network 29

Figure 2-8a

Figure 2-8b

An attempt to process this situation on a computer would cause it to halt and

print out an indication of loop, for the computer would read event 496 as a
precedent to itself (Figure 2.8b). Therefore, in employing computers it is irnpor-
tant to keep track of event numbers used and not used. (Some computer pro-
grams impose the additional restriction that each event number must be larger
than any predecessor number. Such a restriction is cumbersome because it
greatly inhibits network revision and updating; one tends to "run out of num-
bers" or lose track of the numbers used.
Rule 4 is violated when the condition shown in Figure 2-9 occurs. Activities
A and B may be called duplicate activities, since a computer (or tabular method
of computation), using only event numbers for identification, may not be able
to distinguish the two activities, as indicated below:

Network Description Computer Code

Activity A 6-7
Anivity B 6-7
30 1 I Bmic Topior

w
Figure 2-10

One remedy calls for the introduction of a dummy and another event in series
with either activity A or 3 (Figure 2-10). Now the computer can distinguish be-
tween the activities by their different codes.

Network Description Computer Code

Activity A
Activity 8
Dummy

Note that the above solution does not change the logic of the network. Nor
would the logic be changed if the dummy had been placed at the other end of 3,
or at either end of A. If the reader feels that a change in logic has occurred, he
should review the section Interpretation of Rules.
Another way to correct duplicate activities is t o combine them (Figure 2-1 1).
This solution is simple and effective, but it may destroy some of the desired de-
tail in the network; the question of detail is treated in Chapter 3.
Another special restriction for computer analysis is Rule 5. To accommodate
this requirement it is common practice to bring all "loose ends" to a single initial
and a single terminal event in each network, using dummies if necessary. For ex-
ample, one may wish to network a current project that is already past the initial
event. In this case the network would have a number of open ended, parallel
paths at the "time now" point. These loose ends would be connected to a single
initial event by means of dummies, as shown in Figure 2-12.
When methods employing hand computations on the network (or certain
computer programs) are utilized, Rule 5 is not necessary.

A and B

Figure 2-11
 Development of the Network 31

Figure 2-12

USING DUMMIES EFFICIENTLY

While the need for dummies in certain cases has been pointed out, it is preferable
to avoid unnecessary dummies. For example, consider the diagram in Figure
2-13. Evidently, activity D depends on C, B, and A. But the dependency onA
is clear without the dummy 2-4, which is redundant. Such dummies should be
eliminated to avoid cluttering the network and to simplify computations. (Some
computer services base their charges on the number of activities, including
dummies.)
In other cases it may be necessary to introduce dummies for clarity. For ex-
ample, suppose a particular event is considered a milestone in the project, a
point that represents a major measure of progress in the project. This point may
be assigned a scheduled or target date, which may be noted on the network as
shown in Figure 2-14. However, the ambiguity of events, especially merge or
burst events, can cause confusion about the target date notation. Does the date
represent the scheduled completion of activity A , activity B, activitiesA and B,
or the start of activity C? In these cases it may be desirable t o eliminate the pos-
sible misunderstanding by introducing dummies, as in Figure 2-15. In this exam-
ple it is now clear that the date refers only t o the completion of activity A. It is

Figure 2-13
32 I I Basic Topics

Figure 2-14

Figure 2-15

emphasized that this is only an illustration of the flexibility afforded by judicious

use of dummies, and not a required practice at merge points or at milestones.

ACTIVITY DESCRIPTIONS

Thus far in this chapter, most activities in the illustrative diagrams have been de-
scribed by letter codes on the arrows. This has been done both for convenience
and to emphasize the logic of the network representations. In practice it is
much more common to print several descriptive words on the arrow. This avoids
the need for cross-reference with a separate list of activity descriptors.
The descriptions themselves must be unambiguous; they must mean the same
thing to the project manager, the field superintendent, the various subcontrac-
tors, and others expected to use the network. Descriptions should also be brief
and, where possible, should make use of quantitative measures or reference
points. Examples are shown in the following sample applications.

SAMPLE APPLICATIONS

A Machinery Installation

Consider a project involving the installation of a new machine and training the
operator. Assume that the training of the operator can begin as soon as he is
 Development of the Network 33

0
INSTALL MACHINE INSPECT MACHINE

T R A I N OPERATOR
Figure 2-16

hired and the machine is installed. The training is to start immediately after in-
stallation and is not to be delayed for inspection of the machine. The inspection
is to be made after the installation is complete. One might attempt to network
this project as shown in Figure 2-16.
However, this network says that the inspection cannot begin until the opera-
tor is hired, which is a false dependency. To correct this representation in the
network a dummy is added, as shown in Figure 2-17.

INSTALL MACHINE INSPECT MACHINE

HlRE OPERATOR

Figure 2-17

A Market Survey

Consider now a project to prepare and conduct a market survey. Assume that
the project will begin by planning the survey. After the plan is completed, data
collection personnel may be hired, and the survey questionnaire may be de-
signed. After the personnel have been hired and the questionnaire designed, the
personnel may be trained in the use of the questionnaire. Once the question-
naire has been designed, the design staff can select the households to be surveyed.
Also, after the questionnaire has been designed it may be printed in volume
for use in the survey. After the households have been selected, the personnel
trained, and the questionnaires printed, the survey can begin. When the survey is
complete the results may be analyzed. This project may be networked as shown
in Figure 2-18. Note that dummy 4-3 is essential, whereas dummy 6-5 is neces-
sary only if a computer is to be used.
 questionaire

Figure 2-18

Development of a New Product

The network in Figure 2-19 represents a plan for the development and marketing
of a new product, in this case a new computer program.' Note that the dummy
5-7 is used to conform t o Rule 4. The dummy 2-4, however, is used to show
that activities 4-5 and 4-7 depend on activities 1-2 and 3-4, while activity 2-6 de-
pends only on activity 1-2. Another dummy is shown between events 0 and 1;
this dummy is not technically necessary and could have been omitted by com-
bining events 0 and 1. Note also that this network has two critical paths, de-
noted by the heavy activity lines.

NATURAL AND RESOURCE DEPENDENCIES

The network in Figure 2-18 illustrates the fact that there are two types of activ-
ity dependency. Note that most of the activity dependencies are caused by the
nature of the activities themselves; for example, personnel cannot be trained
until they have been hued, and the questionnaires cannot be used in the survey
until they have been printed. Such dependencies among activities may be called
- natural, and this is the most common type of dependency.
Also note, however, that the selection of households is dependent upon the
design of the questionnaire, but only because one group of people is assigned t o
do both jobs (the "design staff' in the project description). This staffing limita-
tion, and the implication that the&@ staff could not do both iobs s@@
neously causes the two activities to be drawn in series (dependent) rather than
' i n parallel. A dependency of this type is not "'natural," but is caused by the
resource limitation. Thus, it may be called a resource dependewy. The re-
sources involved may be personnel, machinery, facilities, funds, or other types of
resources.
Usually it is best to include in the first network draft all resource dependen-
cies that are known and firmly established as ground rules for the project. These
Develo~rnentof the Network 35
firm dependencies represent significant factors in the planning phase of a project
and will often have major effects on the network and resulting schedule. How-
ever, if there is doubt about the number of resources available or how they
should be allocated among the activities, then resource dependencies should be
omitted in the first draft of the network, which would then be based only on
natural logic. In these situations the techniques of resource allocation discussed
in Chapter 7 should be employed.

TIME-SCALED NETWORKS

If a network is plotted on a time scale, similar t o a bar chart, the arrows become
vectors that represent time durations as well as project elements. Tie-scaled
networks have an advantage in p~esentingthe schedule for the project to maE-
agement personnel or others not familiar with networks. In effect, time-scaled
networks are merely extensions of bar charts.
An example of a time-scaled network is given in Figure 2-20. Compare it with
the same project network in Figure 2-18. Note that in the time-scaled version-
dashed lines are used not only to represent a dummy but also represent exten-
sions of arrows beyond their estimated time durations. Thus the dashed-line ex-
gnsions represent "float time" on that path,
-.-
which
-
will be further explained in
Chapter 4.
However, time-scaled networks also have some key disadvantages. It is the
nature of projects to change schedules frequently, almost constantly. Maintain-
ing accurate time-s-densome, especially w h e x
pencil-and-paper drawing methods are used. At each updating of the project

Train
Hire personnel personnel

/ I
questionnaire
Print
3
questionnaire
1

I I I I I I I I 1 I I
0 1 2 3 4 5 6 7 8 9 10
Time in weeks

Figure 2-20 Time-scaled network of sunrev project.

 Devetopment of the Network 37

schedule many future activities have to be redrawn. With networks containing

hundreds or thousands of activities, time-scaling becomes impractical. This diffi-
culty can be overcome on smaller networks by the use of special mechanical
charting devices. In general, though, time-scaled networks are worthwhile only
for occasional presentations for upper levels of management. (See Chapter 6 for
further discussion.)

ACTIVITY-ON-NODE SCHEMES

The graphical method of drawine utilized in this chapter is c w

avow scheme in this text. This scheme was introduced in the original publica-
tions on the Critical Path Method. Since then the arrow scheme has remained
one of the most common means of portraying networks. Many or perhaps most
computer programs for CPM and PERT are designed to accept the predecessor-
successor (I- J ) event code for activities that are associated with the arrow
scheme.
However, the arrow scheme is by no means the only networking procedure,
nor is it the most efficient one if we judge efficiency by the number of symbols
required to portray a given number of activities in a network. At least three
other schemes have been conceived and used widely. In this chapter other
schemes are described and evaluated. The individual or organization that has a
choice of graphic schemes (that is, not limited by contract or customer prefer-
ence) should consider all of them and select the one most suitable t o his pur-
poses. In most cases, of course, it will be desirable that the organization stan-
dardize its networks by using only one of the schemes consistently.

BASIC NODE SCHEME

7,

The complete reverse of the arrow scheme is the node sc- in which the
nodes represent the activities and the arrows are merely connectors, denoting
precedence relationships. The market survey project in the node format is illus-
~- ~

trated in Figure 2-21. The principal advantage of the node scheme is that it
eliminates the need for special dummies to correct false dependencies. This fea-
ture makes the scheme more efficient and, more importantly, easier to learn. In
the arrow and event schemes the most difficult aspect is learning to make the
proper use of dummies. In the node scheme all the arrows are dummies, in ef-
fect, and there are no subtle false dependency problems requiring the use of spe-
cial dummies.
One of the first proponents of the node scheme and, apparently, its originator
was J. W. Fondahl of Stanford University. Professor Fondahl developed a node
format in 1958, almost concurrent with the publication of the first PERT and
CPM reports (see history of networking methods and references in Chapter 1).
38 l 1 Basic Topics

Figure 2-21 Suwey network, node scheme (shape of node symbol is not significant).

However, the PERT and CPM reports received the earliest public attention and
the arrow format became the most popular scheme during the first decade of
network applications. Most of the computer programs for network analysis
would accept only the arrow (1-4 notation and logic. Later, in the mid-
seventies, commercial computer software that would accept the node scheme
began to appear. Since then there has been a noticeable shift in popularity to-
ward the node scheme. An informal survey by the authors in 1981 found that
users of networking methods at that time were about equally divided between
the arrow and node schemes.

PRECEDENCE DIAGRAMMING

An extension of the basic node scheme was first mentioned in Fondahl's 1961
report (Chapter 1, Reference 14), in which "lag" values associated with activity
relationships and "precedence" matrices were introduced. These concepts
gained further notice in the user's manual for an IBM 1440 computer program
for network processing, published around 1964. One of the principal authors of
that manual was J. David Craig.
As an example of how the precedence rules apply, consider the project net-
work in Figure 2-22. Upon reviewing this f ~ s draft
t of the network, let us as-
sume that the management is dissatisfied with the overall duration of 44 weeks
and seeks ways to shorten the time. They find that the times for drawings and
procurement are dictated by some complex iron castings, and that the other
 h w l o p m e n t of the Network 39

Drawings Procurement Fabrication Assembly

21 5 2
Figure 2-22 First draft o f network. Total time is 44 weeks.

Design Design Hdwe Procure Fabricate

other hdwe ,,-, drawings hdwe ,-., hdwe
2 u 3 13

Procure Fabricate
castings Assemble
3 2
Figure 2 - 2 3 Revised network with split activities. Total time is reduced t o 37 weeks.

hardware requires less time to design, prepare drawings, and procure. Thus, all
of the activities except the last one (assembly) may be split to form two essen-
tially parallel paths for the castings and other hardware. Figure 2-23 shows this
logic, resulting in a total project time estimate of 37 weeks. The revised network
in Figure 2-23 is converted to the node scheme in Figure 2-24.
However, assume that under further study it is recognized that the activities
are still not as discrete and separate as implied by the revised networks. In par-
ticular, management notes that the start of drawings on the castings does not
have to wait until all of the castings are designed (4 weeks). Actually the first
couple of castings can be designed in the first week and the drawing work can
begin with these. Likewise, some of the hardware procurement can begin after
the first two weeks of drawing work on the hardware. These facts can affect the
overall project schedule and it may be worthwhile to revise the network still fur-
ther in order to display the more detailed logic. (As an exercise, the reader may
sketch such a further revision, using the node scheme.)
In order to represent "overlapping" activities of the kind occurring in this ex-
ample, modifications of the node scheme were developed. The modifications
consist of defining precedence relationships among activities, which are illus-
trated in Figure 2-25. In addition, one may specify a "lag time" associated with

-
Design Design Hdwe Procure Fab.
casting$ other hdwe ---c dwgs hdwe hadwe
4 2 13 2
3
-
\ \ .

Castings Procure Assemble

2
dwgs castings
7 21

Figure 2-24 Revised network- node schemq. Total time is 37 weeks.

40 l l Basic Topics

Figure 2-25a Finish-to-stan relationship. (Start of B must lag 5 days after the finish of A.)

OR m-7 ss=3

Figure 2-25b Start-to-stan relationship. (Start of B must lag 3 days after the start of A)

Figure 2-2- Finish-to-finish relationship. (Finish of B must lag 3 days after the finish of
A.1

Figure 2.25d Stan-to-finish relationship. IFinish of B must lag 45 days after the start of
A.1

Figure 2-25e Composite start-to-stan and finish-to-finish relationships. (Start of B must

lag 2 days after the start of A, and the finish of B must lag 2 days after the finish of A.1

any of the precedence relationships. which can be used to account for the over-
lapping times among activities.
The first relationship in Figure 2-25a is the only one allowed in PERT and
CPM, with the additional flexibility that a lag time is allowed. For example, the
start of "concrete forms stripping" (activity B), must lag 5 days after thefinish
of "pour concrete" (activity A). The lag of 5 days is required in this case to
allow the concrete to cure and strengthen before the forms are removed. (Whine
the lag times associated with the finish-start constraint are usually zero or posi-
tive, it is possible to use a negative time to allow activity B to start a specified
number of time units before activity A is finished. However, this same result can
 Development of the Network 41

be achieved without the use of negative lags by using the next type of relation-
ship, start-to-start.)
The second relationship, start-to-start, is also used quite frequently. Suppose
activities A and B are "level ground for concrete slab," and "place mechanical
pipe," respectively. Then the diagram in Figure 2-25b indicates that "place me-
chanical pipe" may start 3 days after "level ground for concrete slab" starts.
The 3 day lag here is to allow the grading operation to get a convenient distance
ahead of the piping, which is following behind. Now, if the grading should be in-
terrupted for some reason, it is understood that the piping activity would also be
interrupted. This type of dependency is not shown explicitly in precedence
diagramming, but this does not pose a real problem because field supervisors are
adept at handling this type of situation without the aid of project networks.
The third relationship, finish-to-finish, is also frequently used. For example,
suppose activity A is "place electrical" (5 days) and activity B is "complete
walls" (15 days). Note that it takes 3 days of activity B work to complete the
output from one day of activity A. Therefore thefinish of "complete walls-B"
must lag behind the jinish of "place electrical-A" by 3 days because it will take
3 days of wall completion work to handle the last day of electrical work. This
relationship is shown in Figure 2-25c.
The fourth relationship shown in Figure 2-25d, start-to-finish, is not fre-
quently used. An example will be taken from the field of automative design to
illustrate that precedence diagramming is not restricted to construction work.
Suppose activity A is "design power train" (40 days), and activity B is "design
chassis" (30 days). The dependency relationship between A and B might be that
the last 20 days of the chassis design work depends upon the results that will be
obtained from the first 25 days of power train design work. Thus, 25 days after
the start of the power train, 20 days is required to finish the chassis design. This
implies that the finish of chassis design must lag 25 + 20 = 45 days after the start of
the power train design. This is shown in Figure 2-25d by the SF = 45 relationship.
The composite relationship shown in Figure 2-25e is required wherever a se-
ries of activities must follow each other. For example, consider the following se-
quential construction tasks:

A-Erect wall frames

B-Place electrical lines
C-Attach wall boards
D-finish seams on wall boards

Activity B cannot start until 2 days after A has started, and similarly, B cannot
fmish until 2 days after A has finished. This relationship is shown specifically in
Figure 2-25e. The lag of 2 days is chosen from experience as a reasonable sepa-
ration between the carpenters and electricians. If all four of these tasks work at
the same rate (complete same number of walls per day), then this same relation-
 42 I / Basic Topics

Design
Hdwe Procure Fabricate
castings
dwgs hdwe hdwe
&.hdwe
3 13 2
6
I, 0 b
L=2 L =1
Assemble
4 2

Castings Procure Fabricate

L= 1
c dwgs -a castings -I castings
7 21 3
B
L=2

Figure 2-26 Sample project converted to precedence diagramming network. Total time i s
reduced to 33 weeks.

ship would hold between activities B and C and between C and D. It is not diffi-
cult to visualize the savings in numbers of activities possible with precedence
diagramming in networking a construction type of project. Also, it will be seen
that representing a task by a single activity, rather than a series of activities, sim-
plifies project time and cost control.
Using these "precedence diagramming" rules in revising the network of Figure
2-24,, the management is able to show the overlapping activities in a more accu-
rate way without splitting the activities further, as shown in Figure 2-26. In
fact, there is one less node as compared with Figure 2-24, and it can be shown
that the project duration is further reduced to 33 weeks.
It is evident that in some situations precedence diagramming can be more effi-
cient than arrow or node schemes. Project managers may also l i e the fact that
certain key activities do not have to be split into two or more separate nodes.
In this example, the design work can be retained as a single activity. Advocates
of precedence diagramming feel that it is thus easier to understand and less con-
fusing than splitting activities. However, the method also introduces some com-
plexities of its own in the form of connecting arrows with several different def-
initions, and project time calculations that are not quite as straightforward as in
arrow or node networks.
The subject of precedence diagramming is covered further in Chapters 3 and 4.

SUMMARY

This chapter has been concerned with the translation of the project plan into a
series of interconnecting activities and events composing a network model of the
plan. A few rules were presented, some being required in order t o maintain ac-
curacy and consistency of network interpretation, and others being required by
the nature of data-processing procedures. The rules presented relate t o the ar-
 Development of the Network 43

row method of networking. The next chapter continues the development of the
network, including the addition of activity time estimates and the attainment of
the desired level of network detail.

REFERENCES

1. GE 225 Application, Critical Path Method Program, Bulletin CPB 198B, Gen-
eral Electric Computer Deprtment, Phoenix, 1962.
2. Fondahl, J. W., A Noncomputer Approach to the Critical Path Method for
the Construction Industry, 2nd Ed., Department of Civil Engineering, Stan-
ford University, Stanford, Calif., 1962.

EXERCISES

1. Review the machinery installation sample network in Figure 2-17 and as-
sume that an activity consisting of "schedule inspector" must precede "in-
spect machine." Add the activity to the network without causing a false de-
pendency.
2. In Figure 2-27 find at least five errors or unnecessary symbols. State which
rule is broken in each case, and suggest how the error might be corrected.

Figure 2-27

3. Given the activities and relationships listed below, draw an accurate network
with no more than six dummies. Check your results by (1) numbering each
event (do not repeat any numbers) and determining that no activity or
44 l 1 Basic Topics

dummy has the same pair of identification numbers, (2) making sure that
there is only one initial and one terminal event.'

Activity Predecessor Activity Predecessor

A I A, 8. C, D
B J 0.E. N
C K B. C, D
D L K
E M 8,c. D
F N 8,C, D
G 0 .D
A. 8. C
H

4. Using the' list of activity dependencies given below, draw an accurate and
economical (minimizing the use of dummies) network.

Activity Depends on Activity Activity Depends on Activity

A none G F
B -A I F
F A J H
H A K I and J
C B L G, D, andE
D 8 M K
E C N L and M

5. Figure 2-28 shows a portion of a network for construction of a multi-story

building. What do the dummies 29-33 and 34-38 represent? Can you find
other dependencies of this type in the network?

Seeand floor structure

Sech"dZ
Flrn floor inferior
~ormcols. Elecmech. Pourcolr Cure&

LOLL2

C3

Figure 2-28
 Development of the Network 45

6. Assume that the following list of activity dependencies is correct. The dia-
gram in Figure 2-29, however, does not represent these dependencies prop-
erly. Correct the diagram using only one dummy activity.

Activity Depends on Activity

A none
B none
C A
D B and G
E C and D
F D
G A
H E

Figure 2-29

7. Draw a network of the following steam pipe maintenance project. The

project begins by moving the required material and equipment t o the site
(5 hours). Then we may erect a scaffold and remove old pipe and valves (3
hours); while this is being done, we may fabricate the new pipe (2 hours).
When the old pipe and valves are removed and the new pipe is fabricated, we
can place the new pipe (4 hours). However. the new valves can be placed ( 1
hour) as soon as the old line is removed. Finally, when everything is in
place, we can weld and insulate the pipe (5 hours).
8. Consider as a project the servicing of a car at a filling station, including such
normal activities as filling the gas tank, checking the oil, etc. List the activi-
ties you wish t o include, then draw a network of the project. Assume that
there are two people available t o senrice the car and that the gas pump has
an automatic shut-off valve.
In carrying out this exercise, remember that the netwrok should reflect
technological or physical constraints rather than arbitrary decisions on the
order in which the activities are t o be carried out. It is therefore suggested
 that you include time estimates for each activity and then redraw your origi-
nal network on a time scale to show the actual order of scheduling each ac-
tivity so as t o minimize the time required for two people t o service a car.
9. A common error in networking is t o place the activities in the network ac-
cording to the order in which you plan t o carry them out, rather than in the
order dictated by technological predecessor-successor relationships. Con-
sider the following network.

The network was drawn this way because the planner felt that he would not
start activity D until B and C were completed, and also that he would not
start activity E until D was completed. If the true technological require-
ments on activity D are that only A must precede D, and no other activity
depends on activity D, how should this network be drawn?
10. Using the basic node scheme, draw the networks described in exercises 3,4,
and 6.
TIME ESTIMATES
A N D LEVEL OF
DETAIL

By applying the networking rules presented in Chapter 2, one may develop the
fust draft of the project network. The basic logic of the plan should be estab-
lished by the fust draft. The next step is to add time estimates to each activity
and refine the network as needed to display the desired level of detail.
In practice, the processes of time estimation and network refinement are
closely interrelated and are usually accomplished at the same time. For as one
begins to make time estimates, it is found that certain activities need t o be re-
defined, condensed into fewer activities, or expanded into more, in order to
represent the project accurately and at the desired level of detail.
As mentioned in Chapter 1, there are two methods of applying time esti-
mates: the single estimate method and the three estimate method. This chapter
wiU treat only the single estimate method. The three estimates and their asso-
ciated statistical treatment are considered in detail in Chapter 9. The three esti-
mate procedure may be useful in projects where the time values for each activity
may be uncertain and the estimator wishes to provide a range of time values.

TIME ESTIMATION IN THE NETWORK CONTEXT

Accuracy of Estimates

One of the most common first reactions of persons being introduced to critical
path methods is that the whole procedure depends upon time estimates made by

47
project personnel. Since these estimates are based upon judgment rather than
any "scientific" procedure, it is argued, the resulting CPM or PERT schedules
cannot be any better than schedules derived from bar charts or any other method.
While it is true that critical path methods depend upon human time estimates,
as all project planning schemes must, there are some significant differences in
how the estimates are obtained and in how they are used. To illustrate these
differences, consider how estimates for project times and costs are usually de-
rived. The process is similar in most types of industrial or construction projects,
so let us take a familiar example, the construction of a house. A builder's esti-
mate for a house may look something like this:
Cost Time
Estimate Estimate
Clearing and grading $ 1,920.00 -
Foundations 5,336.00 -
Framing 7,290.50 -

Approximately
Total $221'373*58 3 to 4 Months

The total cost was developed from detailed, item-by-item estimates. The total
t i e , however, was simply a gross estimate based on the builder's experience
with similar projects. Careful attention to costs, of course, is the home builder's
key to obtaining contracts and making a profit. He can afford to be less accu-
rate about time estimates because the contracts do not normally have time limi-
tations and the owner is not as concerned about the exact duration of the project.
But suppose the time were a critical factor, as it often is in industrial projects,
and suppose the accuracy of the time estimate were made important to the
builder's profit, through penalty clauses and other means. How could the
builder develop a more accurate estimate of the project duration? We would
expect him to break down the job into its time-consuming elements, to obtain
good time estimates of each element, and to sequence the elements into a plan
that would show which elements must be done in series, which in parallel, etc.
In principle we are saying that greater accuracy in time estimation can be de-
veloped in a manner similar to the development of accurate cost estimates-
through detailed, elemental analysis.
Critical path methods provide a disciplined vehicle for making detailed time
estimates, for graphically representing the sequence of project elements, and
for computing schedules for each element. Thus the network format effectively
utilizes more detail than does a bar chart, which is a particular advantage in
the use of time estimates and in schedule analysis. Human knowledge remains
the basic ingredient in all methods, but it is how the knowledge is applied that
determines the accuracy and power of the results.
 Time Estimates and Level of Detail 49

Who Does the Estimating?

Certainly another key factor in the accuracy of time estimates is who makes
them. A general rule in this regard is that the most knowledgeable supervisory
person should estimate each activity. This means, for example, that activities
that are the primary responsibility of the electrical subcontractor should be esti-
mated by the subcontractor's manager or supervisor most familiar with the job;
activities of the research department should be estimated by the research super-
visor responsible for and most familiar with the work; and so on. The objective
in obtaining time estimates should be to get the most realistic estimates possible.
It is characteristic of network planning to call meetings of all supervisory per-
sonnel at the time-estimating stage and to consider each activity for which they
are responsible. (It is also desirable for these personnel to participate in prepara-
tion of the first network draft. This is not always practical, though, if everyone
involved is not familiar with network principles). In addition to the psychologi-
cal advantages mentioned in Chaper 2, the participation of the key members of
the project team has major advantages. Whenever the subcontractors, suppliers,
inspectors representing the customer, etc., meet to discuss a project, the discus
sion will lead to questions of priorities in certain phases of the work, potential
interference of work crews, definitions of assignments of engineers, and many
other details of planning that might not have been explored until problems arose
during the project. These discussions often identify and resolve potential prob-
lems before the project begins-rather than tackling them as they actually occur,
which can result in the corrective action being expensive or perhaps impossible.
Here again the network merely serves as a cause for calling the meeting and
as a detailed agenda. Yet these thorough planning sessions around a network re-
sult in what is probably the major benefit of critical path methods as practiced
to date. This benefit is the project plan itself, in terms of its validity, its com-
prehensive scope, and its efficiency in the utilization of time and resources. This
is not to say the reader should stop here. There is more to be gained, and many
of the more experienced users are applying critical path methods and related
procedures to advance the science of project management in a variety of ways.
However, it appears that a majority of satisfied users of CPM and PERT have
gotten their money's worth out of the initial network planning effort. Follow-
ing through with the technique in the project control applications, whenever it
is done, has provided bonus benefits.'
The question of bias or "padding" of time estimates is, naturally, related t o
who makes the estimates and their motivations. It does not necessarily follow
that the most knowledgeable person is also the most objective. It is human
nature to try to provide a time estimate that will be accepted as reasonable but
will not likely cause embarrassment later. Thus, a certain amount of bias is t o
be expected in any procedure.
 Nevertheless, it is generally felt by CPM users that the network approach
tends to help reduce the bias to a manageable level. Again, the increased de-
tail shown by a network plays a useful role. The smaller the work elements, the
more difficult it is to hide a padded estimate. Indeed, a certain amount of pro-
fessional pride is often noticeable in the estimator's attitude, which leads to a de-
gree of optimism in his figures. Another factor.favoring realistic estimates is the
recognition that biased figures will tend to make the activities involved form the
critical path, thus invoking concentrated attention of management and other
parties engaged in the project on the group responsible for those activities.

Research on Time Estimates

Only a limited amount of research on the subject of network time estimation has
been published. One of the more interesting studies, by Seelig and Rubin,' com-
pared the results of 48 R & D projects, some of which were "PERTed" and some
of which were not. The authors concluded that the use of PERT definitely did
lead to improvement of schedule performance but had no noticeable effect on
technical performance. Furthermore, they concluded that the improvement in
schedule performance was primarily a result of improvement in communication
among the project managers, which was brought about by the use of the net-
work method.

ESTIMATION METHODS

When to Add Time Estimates

It is best to complete a rough draft of the total project network before any time
estimates are added. This procedure is conducive to concentration on the logic
of the activity relationships, which must be accurately established. When the
draft appears to be complete, the time estimates should be added to each activ-
ity. This step will constitute a complete review of the network, and will usually
result in a number of modifications based on the diagrammer's new perspective
of the total project network.
As soon as the estimates are completed, a simple hand computation of the
forward pass should be made (this results in the earliest activity start and finish
times, as explained in the next chapter). This is an important step, for it may
reveal errors or the need for further refinement before the preparation of the
final working draft of the project network.

Conventional Assumptions

The time estimate to be made for each actMty is the mean or average time the
activity should take, and the estimate is called the activity duration. This term
 Time Estimates and Laral of Detail 51

is employed to imply that elapsed time of the activity expressed in units such
as working days, rather than a measure of effort expressed in units such as
man-days. Units other than working days, such as hours or weeks, may be
utilized, provided the unit chosen is used consistently throughout the network.
Estimates of activity duration do not include uncontrollable contingencies such
as fires, floods, strikes, or legal delays. Nor should safety factors be employed
for such contingencies.
In estimating an activity's duration time, the activity should be considered
independently of activities preceding or succeeding it. For example, one should
not say that a particular activity will take longer than usual because the parts
needed for the activity are expected to be delivered late. The delivery should
be a separate activity, for which the time estimate should reflect the realistic
delivery time.
I t is also best to assume a normal level of manpower, equipment, or other re-
sources for each activity. Except for known limitations on resources that cause
some activities to be resource dependent (discussed in Chapter 2), donot attempt
to account for possible conflicts between activities in parallel that may compete
for the same resources. These conflicts will be dealt with later, after the schedul-
ing computations have been made.

Fixed-Time Activities

Certain project activities require fxed time periods beyond the control of man-
agement. Examples include: legal minimums of 30 or 90 days for advance no-
tice for public hearings or other events, advance deadlines of media for receipt
of advertising copy, technical minimums such as the time to cure concrete or to
transport equipment by ship or barge. Often such activities become part of the
critical path and thus are important factors in the total project duration.
What can be done about such fured-time activities? Of course, the most im-
portant step is to recognize these time periods and deadlines early in the plan-
ning phase of the project, so that the preparatory steps are scheduled to avoid
missing the deadlines. Network planning methods help to identify such dead-
lines early and to plan properly to meet them.
Another consideration is to make sure that as much as possible can be ac-
complished in parallel with fmed-time activities rather than in series with them.
This may help reduce the total length of time on the paths containing the fxed
elements. In some cases hard, creative thinking may even uncover ways t o re-
duce or eliminate "fxed-time" activities. For example, a new product advertis-
ing campaign might be restructured to reduce dependence on long lead time
monthly publications, while shifting emphasis to weekly publications, radio
and television spots, or other media that can accept copy on shorter notice.
Considered early enough, even legal and technical limits can be modified some-
52 I 1 Basic Topics

times by redesigning the product or the project procedures, if the schedule im-
peratives warrant such fundamental revisions.

Accounting for the Weather

In construction projects the weather is one of the greatest sources of scheduling

uncertainty. In a single-estimate system, there are two common approaches for
taking the weather into account.
The first approach is t o omit consideration of the weather when estimating
the duration of each activity, and instead, estimate the total effect of weather
on the project's duration. For example, suppose a project's duration is com-
puted to be 200 working days. Consideration is now made of the seasons in
which the outdoor work will be done, the seasonal temperatures and precipita-
tion in the region, the type of soil, type of construction, and other weather-
related factors. I t may be estimated that five weeks would be lost because of
bad weather. Thus the total project duration would be increased t o about 225
working days. However, this approach is no different from the usual method
of accounting for the weather in construction estimating. It does not take ad-
vantage of the detailed breakdown of activities afforded by the project network.
The second approach involves the consideration of weather effects in making
each activity time estimate. In this approach each activity is evaluated as to its
weather sensitivity-excavation work being sensitive to rain, concrete work sensi-
tive to freezing, interior plumbing not weather sensitive, etc. Suppose an ac-
tivity is estimated to require ten mandays, and two men will be placed on the
task; the nominal time estimate will be five working days. The weather sensi-
tivity of this activity, the season, and other weather factors may indicate that
this activity's estimate duration should be increased about 20 per cent. Thus,
the adjusted time estimate is six working days.
The initial scheduling calculations may also provide data useful in dealing with
possible weather delays. Activities with relatively large amounts of float may be
scheduled early or late in order t o avoid bad weather seasons. (The definition
and uses of float are treated in Chapter 4.)
The advantage of this detailed approach to weather adjustments is that it
applies the adjustments to particular portions of the network, which will result
in a more accurate schedule for each activity with reference to calendar dates.
A disadvantage is the need t o add more notations to the activity descriptions.
Activity descriptions, including time estimates, must be clearly understood by
all persons expected to work with them. This means that both nominal and
weather-adjusted estimates should be noted on each activity; in some cases, it
is desirable to add the man-day estimate as well. With these notations, on ac-
tivity may appear as in Figure 3-1, where MD = man days, WD = working days.
A legend on the network should explain the notations, including the fact that
the number standing alone represents the weather-adjusted estimate in working
days
 Time Estimates and Level of Detail 53

Erect scaffold
(IOMD, 5 W D )

Figure 3-1

Accounting for Weekend Activities

The use of working days, which is a common time unit in construction projects,
results in computations of project duration which assume that no activities pro-
ceed on weekends and holidays. However, this may be incorrect. For example,
concrete may be cured and buildings may be dried out over nonworking days.
In such cases, time estimates in working days tend to result in an overestimate
of the project duration. When activities of this type are expected to take longer
than 5 calendar days, the overestimate can be corrected t o a certain extent by
a suitable adjustment of the working-day estimate. For example, a curing re-
quirement of 6 or 7 days can be estimated as 5 working days (assuming a five-
day work week). A curing requirement of 5 days, however, may actually take
5 working days, although it would likely run over a weekend and thusconsume
only 3 or 4 working days; in such instances, the estimator should employ the
project network to judge the likelihood of curing over the weekends and adjust
the working day estimate accordingly.
All time estimates in a network must be based on the same number of work-
ing days per week. For activities that will deviate from this standard, adjustments
must be made in the time estimates similar to the adjustment for curing activities
mentioned above.
Most commercial computer programs for network processing include calendar
dating features which automatically convert working-day schedules into calendar
dates, including allowances for weekends and holidays.

A C T I V I T Y REDEFINITION

As in networking, the proper application of time estimates depends primarily on

judgment and experience with critical path techniques. Illustrated here are some
of the common networking problems uncovered when one is attempting to make
time estimates. In most cases the problems involve activity definitions and the
question of detail. Alternative solutions to each problem are discussed.

Activities in Parallel

In the situation illustrated in Figure 3-2a several interrelated activities may begin
and end at approximately the same nodes.
Suppose that upon supplying the time estimates, howkver, it is realized that A
cannot begin until B has been underway for one day, and will not be completed
until one day after B is completed. Furthermore, Ccan be accomplished at any
54 1 I Basic Topics

Figure 3-Za

time B is going on. A practical illustration of interrelationships of this type is

the work prior to pouring a slab of concrete in a building, where the steel, me-
chanical, and electrical trades may begin and end their work at different times,
but during much of the time they are all working in the same area.
To network situations of this type accurately, using the basic arrow scheme,
we must redefine the activities. We recall that an activity is any portion of the
project that may not begin until other portions are completed. Using the basic
arrow scheme, then, and the conditions prescribed above, we may correct Figure
3-2a as shown in Figure 3-2b.
Note the use of percentages to define activities in this illustration. This is
occasionally a useful device, but frequent use of percentages is not good prac-
tice. They often represent arbitrary definitions which can lead to misunder-
standings by subcontractors and others involved in interpreting the network.
Where possible, it is better to use physical measures, such as yards of concrete
poured, the number of columns formed, the specific items assembled, etc.
Another approach to the correction of the activities-in-parallel problem would
be to condense them into a single activity. However, the single activity should
have a t i e estimate representing the total time for the completion of all three
activities, a time estimate which is most accurately obtained from a detailed solu-

Figure 3-Zb
 Time Estimates and Level of Detail 55

A, 6, and C
I1

Figure 3-22

tion as given in Figure 3-2b. From this figure it is clear that the total time re-
quired for all three activities is 11 days. Thus the condensed activity may be
represented as shown in Figure 3-2c.
Whether 3-2b or 3-2c is the "best" solution to the problem depends on the
project, the network objectives, the areas of responsibility involved, and other
factors which can be resolved only through the judgment of the project manager.
This is essentially the problem of the level of detail, which will be discussed fur-
ther. It should be noted here, however, that accuracy and detail are directly
related, and even when less detail is desired in the final draft of the network, it
is often useful in time estimations to sketch certain portions of the network in
greater detail.
Some project management systems use the condensation method shown in
Figure 3-2c in the computerized model of the project, while the supervisor in
the field monitors the actual progress of the work with a hand-processed net-
work in more detail (of the type shown in Figure 3-2b).
These activities-in-parallel situations, or "overlapping activities," occur fre-
quently in construction projects and in certain other kinds of work. The prece-
dence diagramming method introduced briefly in Chapter 2 is designed especially
for representing overlapping activities and should be considered when these rela-
tionships are expected to occur frequently in the project. To compare the
method, let's first illustrate the sample problem in the activity-on-node format.
See Figure 3-2d. In this case there is not much difference in network efficiency
or clarity; there are stiil five defmed activities.
Now consider the problem represented in a precedence diagram, Figure 3-2e.
(To analyze this network, review the precedence relationships defined and illus-
trated in Figures 2-25 and 2-26.) Note that Figure 3-2e shows only the three

Figure 3-2d
56 l l Basic Topics

Figure 3-20

basic activities, and their overlapping relationships are displayed by means of

the dependency arrows and lag-time notation. The main feature of the prece-
dence notation is that the activities are not subdivided on the network, which
is considered an advantage for readability. However, the dependency relation-
ships are now more involvled. For some people, the precedence diagram may
be more difficult to master. (Comparison of scheduling computations for the
arrow, node, and precedence methods is presented in the next chapter.)

Activities in Series

Let us look at another problem of network accuracy, this one arising in a por-
tion of a network in which the activities are drawn in series, as shown in Figure
3-3a. Suppose that upon inspecting this network it was realized that A and B
did not require a total of 17 days. Actually part of B could begin at least 2
days before A was completed. (This is similar t o the previous problem, since we
are saying that A and B are partly concurrent.) An erroneous diagram of this
type may be corrected in several ways. One way would be to split A at the point
that B begins (Figure 3-3b). Another way would be to simply absorb the com-
pletion of A in B, making sure that the activity descriptors were clear (Figure
3-3c). A third way, of course, would be to condense A andB (Figure 3-3d).

Figure 3 3 s

Begin A Complete A

Figure 3-3b
 Time Estimates and Level of Detail 57

Begin A Band complete A

Figure 3 . 3 ~

A and B
15

Note that in each of the alternative solutions illustrated in Figures 3-3b, 33c,
and 3-3d, the total elapsed time for A and B is shown correctly as 15 days.
Again, the choice of a solution depends on what one wishes t o flustrate and con-
trol, and on such factors as the magnitude of the times involved and the feasi-
bility of defining the activity segments clearly.

Practical Example

The problems discussed above arise repeatedly in practical efforts t o draw accu-
rate networks. Therefore, it is worthwhile to review the points made in the form
of a small practical example. Consider the project of preparing a technical re-
port. The major activities include the original writing, calculations, and chart
sketches done by an engineer, typing and graphics work done by technical sup
port personnel, and fmal printing and binding done in the reproduction depart-
ment. The engineer might first draw a network and add time estimates as shown
in Figure 3-4a.
The author's reaction to this network might be that 31 days is too long. Her
deadline might be only 28 days away. Looking at her report outline, she may

Draft text Typing & Proof & Print &

graphics correct bind
6 3 2
Figure 3-49 First network of report project. Total project duration is 31 days.

Draft main Draft

Proof &
correct ,-, Print & bind
3 w
I I
6
Figure 3-4b Revised n e t w o r k Project duration is 26 days.
note that it divides easily into two parts: (I) the main text and charts, and (2)
the technical appendices. Consulting the typing and graphics departments, she
finds that they can readily handle the report in those two parts. Thus she re-
.
vised the network as shown in Figure 3-4b. The resulting total time of 26 days
allows for 2 days of errors in time estimates or unexpected delays.

NETWORK CONDENSATION

In the foregoing illustrations the concepts of condensing and expanding net-

works are introduced for the purposes of improving accuracy, eliminating exces-
sive detail, and to achieve other objectives in the development of the
detailed network. There are also occasions in which it is desired to produce a
summary network for review by top management, which calls for the same con-
densation concept illustrated above, except that it is applied on a broad basis
throughout the network, with the purpose of developing a general condensation
and summarization of the project plan. In practice this often means that a net-
work of several hundred activities must be reduced t o one of a few dozen activi-
ties, without distortion of the logic, such that a summary picture of the project
may be presented for review by top management, a customer, or other interested
audiences. Consequently, some points related to condensation procedures are
worthy of attention.
In general, a safe rule of condensation is that groups of activities independent
of other activities may be condensed without distorting the network logic. For
example, consider Figure 3-5a. In this network there are three independent ac-
tivity groups that may be condensed, as shown in Figure 3-5b. Note that some

Figure 3-5a

L &M

Figure 3-Sb
 Time Estimates and Level of Detail 59

Figure 3-6s

activities in series have been combined, and in one case two activities in parallel,
L and M, were combined. But all the dependency relationships in the original
diagram still hold. This is the most important in condensing networks, for
it is very easy to introduce false dependencies. (For a practical example, see
Figure 3-9.)
A somewhat different approach t o condensation is used by certain computer
routines that perform this function. This approach calls for the designation of
certain key events in the network which are not to be omited in the-condensa-
tion procedure. Then all direct and indirect restraints (groups of activities) be-
tween each pair of key events are reduced to a single restraint (activity). To il-
lustrate, consider the network of seven activities in Figure 3-6a. The triangles
denote the selected key events. Using the condensation process described above,
this network would be reduced to the two activities shown in Figure 3-6b.*
This particular procedure is vulnerable to the occurrence of "pathological
cases," in which the number of activities is not reduced or may even be in-
creased. Using the procedure on the network in Figure 3-7a, for example, pro-

Figure 3-7s
60 I I Basic Topics

u
Figure 3-7b

duced the network in Figure 3-7b. However, by selecting key events with this
possibility in mind, one can avoid most pathological cases.

THE LEVEL OF DETAIL

Thus far, comments on the problem of the level of detail in a network have
been associated with questions of accuracy and economy of the presentation.
There are many other factors involved in determining the most appropriate level.
In considering any particular activity or group of activities with regard t o ex-
panding, condensing, or eliminating it, the diagrammer may ask himself several
questions to guide his decision:

1. Who will use the network, and what are their interests and span of control?
2. Is it feasible t o expand the activity into more detail?
3. Are there separate skills, facilities, or areas of responsibility involved in the
activity, which could be cause for more detail?
4. Will the accuracy of the logic or the time estimates be affected by more or
less detail?

Clearly, these questions are only guides t o the subjective decision that must
be made in each case. Generally, after working with one or two networks, a
person will develop a sense for the appropriate level of detail.
That there are no firm rules that may be followed in determininglevel of detail
is illustrated by the following hypothetical case. The project is the construction
of a house. If the network rules of Chapter 2 are followed, one could prepare a
complete network, as shown in Figure 3-8a, or one could take a more detailed
approach, as indicated in the portion of the network shown in Figure 3-8b.
These appear to be clear examples of too little and too much detail. But sup-
pose the house is one of a hundred identical ones in a large housing project
Three activities per house would thus result in a network of 300 activities, plus
 Time Estimates and Level of Detail 61

Figure 3-8a

V
Figure 3-8b

other activities for roads, utilities, etc. Such a broad network may be very use-
ful in analyzing the over-all length of the project, the most desirable sequence of
construction, and other problems of general planning. Furthermore, since the
house construction in this case is a matter of mass production, it would be worth-
while to work out the construction schedule for one house in considerable de-
tail, for any bottlenecks in the schedule for one house would cause repeated
delays in all houses. Thus the detailed network treatment for a typical house
might well be justified under these conditions. Under most other conditions the
approaches illustrated above would indeed represent a useless extreme on the
one hand and an expensive, perhaps impractical extreme on the other.

Cylical Networks

In the house building case cited above, it was stated that both detailed and con-
densed networks may be useful if a number of identical houses are to be built.
It may be generally stated that whenever a project involves a number of cycles of
a group of activities, one should consider (1) developing a detailed network of
the group, (2) condensing the detailed network into a summarized version, and
(3) using the condensed network in the cycles that compromise the total project
network. The purposes of the detailed network are to develop an efficient plan
for the group of activities that will be repeated and to derive accurate time esti-
mates for the condensed version. The purpose of the condensed version is net-
work economy, since repetition of the detailed network would be costly in draft-
ing time and would unnecessarily complicate and enlarge the total project
network. Project types to which this principle would apply include multistory
buildings, bridges, pilot production of a group of missiles, and a series of re-
search experiments.
An example of the application of detailed and condensed networks in a multi-
story building is shown in Figure 3-9. Here the contractor worked out the de-
 Excerpt from master network

Detailed network of typical floor

(structural work)
time estimates are in elapsed hours
55?4cal 6Nrnl.
10,"" 10rmr

Figure 3-9 Portion of network for eleven-story building project showing detailed and condensed versions of structural subnetwork. (Courtesy
Floyd D. Traver & Co., Atlanta.)
 Time Estimates and L d of Dotail 63

tailed network using a time unit of hours in order to balance the crews and mini-
mize delays in the structural work on a floor. Then the condensed version was
repeated for each floor of the building in the total project network.
Figure 3-9 also illustrates that large, complex networks are sometimes difficult
to read. This is not a trivial matter, since one of the primary purposes of net-
works is to provide a more readable format for the project plan. An effective ap-
proach to improving network readability is to provide large labels for major sec-
tions of the network. An example is shown in Figure 3.10.

ARROW VERSUS PRECEDENCE METHOD

Up to this point we have demonstrated network methods by using the arrow

scheme as the basic tool, then showing how the node and precedence schemes
compare under certain conditions.
One of the more thorough comparisons of arrow and precedence diagramming
was prepared by Keith C. Crandall of the University of California, Berkeley.
Professor Crandall developed an example which we reproduce here, courtesy of
the author and the publisher.3
The basic problem is to pour three consecutive floor slabs in a multifloor
building. The problem requirements are as follows:

A. Complete the pour of floor slabs for three floors with the following con-
straints:
1. Complete the floors in order so that the crew for each activity is shown
to move from one floor to the next, with the exception of pouring
which is described below.
2. Start by placing bottom reinforcing steel. Once adequately underway
with bottom steel, start mechanical and electrical rough-ins. Place the
top reinforcing mesh once the rough-ins are sufficiently advanced. When
complete aUow one day for inspection and pour preparation. The pour
completes the work on a given floor.
B. Lead/Lag constraints:
la. Mechanical rough-in can start once the bottom reinforcing is two days
underway.
lb. Mechanical rough-in can be completed no sooner than one day after
the bottom reinforcing.
2. Electrical rough-in can start three days after bottom reinforcing, yet
cannot be completed until two days after bottom reinforcing.
3. Top reinforcing mesh can start once mechanical and electrical rough-ins
are two days underway. Top mesh must be completed two days after
the completion of mechanical rough-in and cannot be completed until
the same day as electrical rough-in.
 Time Estimates and Level of Detail 65

4. The pours can be overlapped as long as one day of pour has been com-
pleted on the previous floor.
C. Estimated durations for each floor.
1. Bottom reinforcing 5 days
2. Mechanical rough-in 10 days
3. Electrical rough-in 3 days
4. Top mesh 3 days
5. Delay for inspection 1 day
6 . Pour 2 days

Let us first prepare an arrow network of this problem, taking just the activi-
ties in the first slab. To illustrate the time relationships more clearly in this
case we first use a time-scaled format with a separate arrow for each activity.
To show the dependency relationships we add dummy arrows labeled to corre-
spond with the lead/lag constraints given in the problem. See Figure 3-11. The
fact that some of the dummy arrows must go backward against time reveals that
this time-scaled diagram is not feasible as shown and that several of the activities
will need t o be delayed (shifted to the right on the network).

Time 0 2 4 6 B 10 12 14 16
I I I I I I I I

-.
Bottom
reinforcing0
5
r
I I 1 ..
Mechanical
rough-in
10
( l a ) I (2a),
<
r
-
\
(2b)
k.
r
lib).-.
..- /
/

Electrical /
/ '
rough~in /
3

Top mesh
3
I Logic
1 im~lied
Inspection
1
I Logic
!implied
Pour
2
(4)

+/
To pour 2nd floor

Figure 3-11 First floor slab arrow diagram o n t i m e scale w i t h some arrows going backward
against time.
66 l l Basic Topics

Time 2 4 6 B 10 12 14 16 18
I I I I I I I I I
Bottom
reinforcing
5
-------- --
Mechanical
rough-in
10
I
Electrical
rough-in
-------
3

Top mesh
3
I

Inspection Note: This diagram gives an early

1 start time schedule; the splitting of
electrical rough-in can be avoided by
delaying its start one day; for the
Pour top mesh activity, a 5 day delay
2 is required.

i
To 2nd floor pour
Figure 3-12 F i r s t f l o o r slab a r r o w diagram o n t i m e scale w i t h backward arrows eliminated.
reflecting correct early start schedule (14 real activities plus 11 dummies).

Now, let us draw a feasible arrow network, again on a time scale. Using the
methods described earlier in this chapter, we subdivide the activities as necessary
in order to properly represent the dependency relationships. The result is Figure
3-12. The backward arrows are eliminated and we have a correct early start
schedule. However, the activities have been subdivided to the extent that the
network has become somewhat complex and tedious to read. It contains 14 real
activities and 11 dummies.
Now, let us go back to the original problem statement and network it using
precedence diagramming. Figure 3-13 accomplishes the purpose with 5 activi-
ties and 9 dependency arrows. In this notational scheme the dependency arrows
are coded as follows:
SS = start-to-start constraint
FF = finish-to-finish constraint
FS = finish-to-start delay
The reader should analyze this network for consistency with the problem
statement and with the arrow diagram in Figure 3-12.
Clearly the precedence diagram is easier to read with respect to the activities
 Time Estimates and Level of Detail 67

I Bottom I

Mechanical

I I
I
I
I I
rough-in

I
I
I
I
1 st floor

"1 day for inspection

Figure 3-13 First floor slab precedence diagram. I5 real activities plus 9 dependencies)
conventional dependency artwork.

themselves, as compared with the subdivided activities in the arrow diagram of

Figure 3-12. However, the dependency relationships in the precedence scheme
are more complex and may be more difficult to understand in practice.
In selecting which networking format to employ, one should consider the
types of activity relationships involved in the projects being considered, the peo-
ple who will be reading the networks, and the computational methods available
(covered in Chapters 4 and 11).

SUMMARY

Although the preparation of the network is only the first phase in applications
of critical path methods, many users have reported that the greatest benefits
from the critical path concepts are derived from this phase alone. They felt that
preparing the network caused them to think through the project in a more com-
plete manner than ever before, forcing them to do a more thorough job of ad-
vanced planning. However, a great deal of useful information is included in the
completed network, and the proper processing and utilization of this informa-
tion as described in the following chapters can bring important additional bene-
fits not only to the project manager but also to the subcontractors and all other
groups engaged in the project effort.
68 1 / Basic Topics

REFERENCES

1 . Davis, E. W., "CPM Use in the Top 400 Construction Firms," Journal o f the
Construction Division, ASCE, March 1974.
2. Seelig, W. D., and I. M. Rubin, "The Effects of PERT in R&D Organizations,"
published as a Working Paper of the Research Program on the Management of
Science and Technology, No. 23066, Alfred P. Sloan School of Industrial
Management, Massachusetts Institute of Technology, December 1966.
3. Crandall, Keith C., "Project Planning with Precedence Lead/Lag Factors,"
Project Management Quarterly, June 1973.
4. Mark, E. J., "How Critical Path Method Controls Piping Installation Progress,"
Heating, Piping, and Air Conditioning, September 1963, pp. 121-126.

EXERCISES

1. Redraw the network in Figure 3-4b, using an activity-on-nodescheme. What

advantages and disadvantages of the node scheme do you see in this example?
2. Redraw the detailed network of typical floor shown as part of Figure 3-9,
using an activity-on-node scheme. Then condense the network into a node
format containing 12 or fewer activities. Compare with the condensed ver-
sion in Figure 3-9.
3. Condense the network in Figure 3-14 t o 10 activities or less, without dis-
torting the dependency relationships.
 Time Estimates and Level of Detail 69

Figure 3-15

4. Condense the network in Figure 3-15 without distorting the logic.

5. Considering Chapters 2 and 3 only, and the initial planning phase of projects
(prior to scheduling computations), what planning techniques can reason-
ably be used t o shorten the total project duration?
6. Considering only the material covered in Chapters 2 and 3, list the ways in
which network methods can be used to improve communications among
project personnel, including management, s u p e ~ s o r s subcontractors,
, sup-
pliers, and the owner or end-user of the project results.
7. Some persons who supervise part of the work in a project, such as depart-
ment heads or subcontractors, may inflate their time estimates somewhat
in order t o more easily stay on schedule and reduce attention and pressure
by the general manager of the project. How will theuse of network methods
and scheduling computations tend t o cause this tactic to boomerang?
8. The network shown in Figure 3-16 is an early application of critical path
methods by the Pure Oil Company, which was published in 1 9 6 3 . ~Review
the network and answer the following questions:
a. Assuming a computer was used to process this network, could any of
the dummies be eliminated without distorting the logic of the network?
b. If hand computations are used, which dummies could be eliminated?
c. What do you suppose the diagrammer accomplishes by the use of the
"activity reference number"?
d. Note that the delivery of certain items, such as truck rack equipment
(activities 11-27, 11-28), are included in the network. Why do you sup-
pose the diagrammer did not include the delivery of all items, such as
concrete, pumps YP-1 and YP-2 (activities 13-16, 14-15), and insulation
(activities 36-37, 77-78), etc?
e. How could the readability of this network be improved?
70 l l Basic Topics
 Time Estimates and Level of Detail 71

9. Discuss how the level of detail used in a network may be influenced by the
purpose and readership of the network. Consider the viewpoints of differ-
ent levels of management, such as: the project superintendent for a high-rise
office building; the president of a television manufacturing firm that is
about to introduce a new unit for the retail market; a subcontractor for
computer programming in a two-year project to develop a management in-
formation system for a major bank.
10. Figures 3-12 and 3-13 are networks of the f i s t slab of construction problem
defined by Crandall. Extend the network to include the second and thud
slabs, using (a) the arrow method and (b) procedence diagamming.
BASIC SCHEDULING
COMPUTATIONS

At this stage in the application of critical path methods, the project network
plan has been completed and the mean performance times have been estimated
for each activity. We now consider the questions of how long the project is ex-
pected to take and when each activity may be scheduled. Answers to these ques-
tions are inferred from the network logic and the estimated durations of the
individual activities. These estimates may be based on a single time value, as de-
scribed in Chapter 3, which is basically the original CPM procedure, or it may be
based on a system of three time estimates, as described in Chapter 9, which deals
with PERT, the statistical approach to project planning. Regardless of which
estimation procedure is used, the scheduling computations described in this
chapter are the same, since they deal only with the estimates of the mean activ-
ity duration time. Since network logic is the underlying basis of these computa-
tions, different procedures are required for arrow and precedence diagrams. It
will be seen that the more extensive nature of precedence diagramming logic
leads to a more involved computational procedure. An understanding of these
basic scheduling computations is essential to proper interpretation of computer
outputs, and also to enable one to carry them out by hand when the occasion
demands. Readers interested in a more rigorous algorithmic formulation of
these computational procedures, essential to writing computer programs, are re-
ferred to Appendix 4-1 at the end of this chapter.
A11 basic scheduling computations first involve a forward and then a backward
pass through the network. Bosed on a specified project start time, the --
-- forward
 Basic SchedulingComputations 73

-
pass computations proceed sequentially from the beginning to the end of the
project giving the earliest (expected) start andfinish times for each activity, and,
for arrow diagrams, the earliest (ex~ected) occurrence time for each event. The
modifier "expected" is sometimes used to remind the reader that these are esti-
mated average occurrence times. The actual times, known only after the various
activities are completed, may differ from these expected times because of devia-
tions in the actual and esfimated activity performance times.
By the specification of the latest allowable occurrence time for the comple-
tion of the project, the backward pass computations also proceed sequentially
from the end to the beginning of the project. They give the latest allowable start
and finish times for each activity, and, for arrow diagrams, the latest aZIowabZe
occurrence time %each event. After the forward and backwak pass computa-
tions are completed, the float (or slack) can be computed for each activity, and
the critical and subcri?ical -~ujhenetwt-rk-dete~-%ed.
- As mentioned in Chapter 3, it is often appropriate to adopt one working day
as the unit of time, so that the network computations are made in working days,
beginning with zero as the starting time of the initial project event. The conver-
sion of these computational results t o calendar dates merely requires the modifi-
cation of a calendar wherein the working days are numbered consecutively from
a prescribed calendar date for the start of the project. This procedure is dis-
cussed further in Chapter 11. For convenience, this chapter will use elapsed
working days for discussion purposes; it should be understood, of course, that
other time units than working days may be used with no changes in the compu-
tation procedures. In addition, it is assumed at the start that the project begins
at time zero and has only one initial and terminal event. These assumptions will
be relaxed later in this chapter.
The basic scheduling computations are also termed "time-only" procedures
because resource limitations are not explicitly considered, unless of course, they
are built into the network logic. The effect of resource limitations on actual ac-
tivity schedules will be considered in Chapter 7 which deals with this specific
problem.

COMPUTATION NOMENCLATURE-ARROW DIAGRAMS

The following nomenclature will be used in the formulas and discussion which
describe the various scheduling computations; for brevity, the modifier "ex-
pected" has been omitted from all of these definitions of time and float. Also,
these definitions and subsequent formulas will be given in terms of an arbitrary
activity designed as (i - j), i.e., an activity with predecessor event i, and successor
event j.
Dij = estimate of the mean duration time for activity (i - j)
Ei= earliest occurrence time for event i
74 l l Basic Topics

Li =latest allowable occurrence time for event i

EXij = earliest start time for activity (i - J>
EFij = earliest fmish time for activity (i - j)
LSij =latest allowable start time for activity (i - j)
LFij =latest allowable finish time for activity (i - j)
Fij = total path float (or slack) time for activity (i - j)
AFii =activity (free) float (or slack) time for activity (i - j)
Ts = scheduled time for the completion of a project or the occurrence of
certain key events in a project.

FORWARD PASS COMPUTATIONS

To compute the earliest start andfinish times for each activity in the project, the
forcvmd pass computations are initiated by assigning an arbitrary earliest start
time to the (single) initial project event. A value of zero is usually used for this
start time since subsequent earliest times can then be interpreted as the project
duration up to the point in question. The computations then proceed by assum-
ing that each activity starts as soon as possible, i.e., as soon as all of its predeces-
sor activities are completed. These rules are summarized below.

Forward Pass Rules-Computation of Early Start and Finish Times

R U LE 1 . The initial project event is assumed to occur at time zero. Letting the
initial event be denoted by I , this can be written as:

RULE 2 . All activities are assumed to start as soon as possible, that is, as soon as
all of their predecessor activities are completed. For an arbitrary activ-
ity (i - 3 this can be written as:
ESij = Maximum of EF's of activities immediately preceding activity
(i - j), i.e., all activities ending at node i
RULE 3. The early finish time of an activity is merely the sum of its early start
time and the estimated activity duration. For an arbitrary activity
(i - j)this can be written as:
EF.. = ES.. + D..
ll ll Y

The above rules are applied to the simple network shown in Figure 4-la. In
the forward pass section, Figure 4-lb, the initial project event 1 is placed at 0 on
the time scale according to the first rule. Starting with El = 0, the early start
time of activity 1-2 is 0, and the early finish time by Rule 3 is merely
 Basic Scheduling computations 75

la) Basic network with activity duration times

Elapsed working days

Figure 4-1 Example of forward and backward pass calculations for a simple network.

The early start and finish times of activities 2-3,3-5,2-4,and 4-5 are determined
similarly, in that order. Proper sequencing of the activities so that the EF's of
all predecessor activities will be available when computing ES's of any follower
activity (use of Rule 2) is easily handled in hand computation on networks by
working along each path as far as possible, and then "back tracking" t o a new
76 l l Basic Topics

path that is then ready for computation. Special procedures required for tabular
computations, or the development of computer algorithms is taken up in Ap-
pendix 4-1.
The crux of the forward uass computations occurs at the merge event 5,
where it is necessary t o consider the early finish times for predecessor activities
3-5 and 4-5 to determine the early start time for activity 5-6, i.e.,
ES,,, =Maximum of (EF,,, = 9 and EF,,, = 7)= 9
Finally, the early finish time of the final network activity is

Thus, the early expected finish time for the entire project, corresponding t o
the earliest occurrence time of the project terminal event 6, is denoted by E, =
EF,,, = 12.
The forward pass network in Figure 4-lb has been drawn to scale on a time
base, not only as a convenient means of showing the earliest start and finish
times for each activity, but also to show the longest path through the network.
This graphical procedure is introduced here for illustrative purposes, but is not
practical for routine hand application. In Figure 4-lb, activities 1-2-3-5-6form
the longest path of 2 t 4 t 3 + 3 = 12 days duration. The path consisting of ac-
tivities 2-4-5 has two days of total path float, as will be discussed below; this
path float is indicated by the dashed portion of the activity 4-5 arrow.

BACKWARD PASS COMPUTATIONS

The purpose of the backward pass is to compute the latest allowable start and
finish times for each activity. These conputations are precisely a "mirror image"
of the forward pass computations. First, the term "latest allowable" is used in
the sense that the project terminal event must occur on or before some arbi-
trarily scheduled time, which will be denoted by Ts. Thus, the backward pass
computations are initiated by specifying a value for Ts. If no scheduled date for
the completion of the project is specified, then the convention of setting the
latest allowable time for the terminal event equal to its earliest time, determined
in the forward pass computation, is usually followed, i.e., L = E for the termi-
nal event of the project. This was followed in the initial development of CPM,
and will henceforth be referred to as t h e m k convention, o ~ e r o - f r o a ?
convention.
One result of using this convention is that the float along the critical path(s) is
zero, while the float along all other paths is positive. When an arbitrary sched-
uled date is used for the project terminal event, the float along the critical path
may be positive, zero, or negative, depending on whether Ts is greater than,
equal to, or less than, respectively, the earliest occurrence time for the terminal
 Basic Scheduling Computations 77

event. The zero-float convention has an additional useful property in that the
Intest allowable activity finish timegives the time to which the completion of an
activity can be delayed without directly causing any increase in the total time to
complete the project.
The zero-float convention is adopted in the illustrative example shown in Fig-
ure 4-lc, where the terminal event 6 is placed at time 12, i.e.,L, = E 6 = 12. The
latest allowable finish time for activities other than the final activity(s) are then
determined from network logic, which dictates that an activity must be com-
pleted before its successor activities are started. Thus, the latest allowable fmish
time for an activity is the smallest, or earliest, of the latest allowable start times
of its successor activities. Finally, the latest allowable start time for an activity
is merely its latest allowable finish time minus its duration time. These rules are
summarized below.

Backward Pass Rules-Computation of Latest Allowable

Start and Finish Times

RULE 1. The latest allowable fmish time for the project terminal event ( t )is set
equal t o either an arbitrary scheduled completion time for the project,
Ts, or else equal t o its earliest occurrence time computed in the for-
ward pass computations.

RULE 2. The latest allowable finish time for an arbitrary activity (i - j ) is equal
to the smallest, or earliest, of the latest allowable start times of its suc-
cessor activities.
LFij = Minimum of LS's of activities directly following
activity (i - 12
RULE 3. The latest allowable start time for an arbitrary activity (i - 12 is merely
its latest allowable finish time minus the estimated activity duration
time.
LSij = LFij - Dii

These rules are applied in Figure 4-lc labeled Backward Pass. Statiig with
the final project event 6, we see that according to the zero-float convention, it is
placed at time 12 which is the earliest time for event 6 computed in the forward
pass calculations.
We next compute the latest allowable start time of activity 5-6 by applying
rule 3,i.e.,LS5,, =LF,,, - D,,, = 1 2 - 3 = 9 . Thecruxofbackwardpasscom-
putations occurs at the burst event 2. Here, the computation of the latest allow-
able finish time of activity 1-2 requires consideration of its two successor activi-
78 1 I Basic Topics

ties. Applying rule 2 above we obtain

LF,,, =Minimum of (LS,,, = 2 and LS,,, = 4) = 2

Finally,we obtainLS,,, =LF,,, - D l , , = 2 - 2 = 0 . Thisresultcanbeusedas

a check on the computations when the zero-float convention is followed. If
L6 =E6 = 12 for the terminal event, then L , = E , = 0 must result for the initial
event.

DEFINITION AND INTERPRETATION OF FLOAT (SLACK)

Among the many types of float defmed in the literature, two are of most value
and are stressed in this text; they are called total path float, or simply path float,
and activity free float, or simply activity float. Path float and activity float are
also referred to by some authors as total slack and free slack, their definitions
being identical to those given below.

Path Float-Total Float

Definition:

&h float, as the name implies, is the totalfloat associated with a path. For
a particular path activity, say ij, it is equal t o the difference between its earliest
and latest allowable start or finish times. Thus, for activity (i - j), the path float
is given by

Assume the zero-float convention is followed, i.e., let L t = E t for the terminal
event. Then the path float denotes the amount of time (number of working
days) by which the actual completion time of an activity on the path in question
can exceed its earliest completion time without affecting the earliest start or
occurrence time of any activity or event on the network critical path. This is
equivalent t o not causing any delay in the completion of the project. For exam-
ple, in Figure 4-lc the path float for activities 2-4 and 4-5 is two days. Thus, the
slack path (2-4-5) is two days away from becoming critical. Now, suppose activ-
ity 2-4 "slips" by starting a day late or taking a day longer to complete than
originally estimated. The result is that its completion time occurs at the end of
the fifth day instead of the fourth day, and the path float for the remainder of
this slack path is thereby reduced by one day. That is, the path float for activity
4-5 is reduced from two to one day. But note that this slippage does not effect
the earliest times for any critical path activity or event. Also note that if the
total slippage along the slack path 2-4-5 exceeds its path float of two days,
 Basic Scheduling Computations 79

then the critical path is affected, and the duration of the project is increased
accordingly.
There is a subtle difference in the interpretation ofpathfloat when the zero-
float convention of letting L , = E, for the terminal event is not followed. Con-
sider Figure 4-ld where E , = 12 but L , = Ts= 14. Here, activity 2-4 has a path
float of four days; it can slip up t o this amount without causing the project com-
pletion to exceed its scheduled time of 14 days. However, a slippage of four
days will cause the early start time of the critical path event 5 to slip by two
days, and hence delay the completion of the project, but only up to its sched-
uled completion time of 14.

Activity Float-Free Float

Merge point activities that are the last activity on a slack path (activity 4-5 in
Fig. 4-1) have what is called activity float, sometimes called free float. The
name follows from the fact that the specific activity is free to use this float with-
out effecting any other activity times in the network. This float concept is not
widely used, but it will be defined below and discussed briefly.

Definition:

A c t i v i t y f l o r t time of the activity's successor

activity(s) minus the earliest finish time of the activity in question. Thus, for
=ty (i - d, the activity float is given as follows, where j - k denotes a succes-
sor to the activity in question.

Activity float is equal t o the amount of time that the activity completion time
can be delayed without affecting - the
.- earliest start or occurrence time of any
other activity or event in the network.
-
In a sense, activity float is "owned" by an individual activity, whereas path or
total float is shared by an activities along a slack path,
Again consider the slack path 2-4-5 in Figure 4-1, and assume the zero-float
convention is used. The last activity on this path, 4-5, has both apath float and
an activity g o a t of two days. If it slips by an amount of up t o two days, no
other network activity or event times are affected. However, this is not true for
activity 2-4 which has no activity flont. Here, any slippage is immediately re-
flected in a corresponding delay in the early start time of its successor activity,
4-5. Thus, when we say that activity 4-5 has an activity float and a path float of
two days, we are assuming that its predecessor will be completed by its emly
finish time.
80 1 1 Basic Topics

CRITICAL PATH IDENTIFICATION

Now that the concept of float has been described, the critical path through a
network will be formally defined as follows.

Definition:

The critical path is the one with the least path float. If the zero-float conven-
tion o f letting Lt = E, for the terminal network event is followed, the critical
path will have zero float; otherwise, the float on the critical path may be positive
or negative. If the network has single initial and teminal events and no sched-
uled times are imposed on intermediate network events, then the critical path is
also the longest path through the network.
For the network in Figure 4-lc, the critical path, the one with least path float,
is 1-2-3-5-6. It is also the longest path through the network in this case. Its du-
ration is equal to 12 days, and it has zero path float since the convention of let-
ting L6 = E, = 12 was followed.
To illustrate the case where the zero float convention is not followed, it has
been assumed in Figure 4-ld that the scheduled completion time of the project
is 14 working days, i.e., L6 = Ts = 14. In this case, we see that the critical path
remains the same, i.e., 1-2-3-5-6, since it is still the path with least path float.
However, the float along the critical path is now positive, that is, two days, while
the float along the slack path 2-4-5 is now four days. In this case, a slippage up
to two days along the critical path will cause the critical path events t o slip a
corresponding amount. However, the critical path activities will not slip beyond
their latest allowable start and finish times, and in particular, the project end
event 6 will not slip past its scheduled completion time of Ts= 14 days.

OTHER TYPES OF FLOAT

Battersbyl defines a third type of float called interferingfloat, as the pathflwt

minus the activity f i a t , or total float minus free float. For an arbitrary activity
(i - 13, this can be written as:

For Figure 4-lc, the activities on the slack path (2-4-5) have the float values
shown in Table 4-1. The interpretation follows from the name. Activity 2-4 has
both path and interfering float of two days; if it uses its float, it "interferes," by
this amount, with the early times for the down path activity, 4-5. However, ac-
tivity 4-5 has no interfering float, but rather it has both path float and activity
(free) float of two days (assuming, of course, that activity 2-4 is completed at its
early finish time). Activity 4-5 can use itsactivity float of two days without "in-
terfering" withany other activity or event times in the network. .
 Basic Scheduling Computations 81

Table 4-1. Float Values for Slack Path

(2-4-5) in Figure 4-lc

Float Type

Activity Path Activiw Interfering

2-4 2 0 2
4-5 2 2 0

Battersby also defines independent float, which occurs only rarely in typical
networks. It is the amount of float which an activity will always possess no mat-
ter how early or late it or its predecessors and successors are. It is found by as-
suming a worst case situation. That is, all predecessors end as late as possible
and successors start as early as possible. If this time interval exceeds the dura-
tion of the activity in question, the excess is called independent float. It is illus-
trated in Figure 4-2, where it can be seen that activity 2-4 has an independent

la)

Early start schedule

Path float = activ~tyfloat = 4
For activlty 2-4

Path float = L, - EF,-, = 8 - 4 = 4

A n ~ v i t yfloat = E, - EF,_, = 8 - 4 = 4
lnterferlng float = 4 - 4 = 0
Succeswr 14-51 ES = 8
(b) Predecessor (1-2) LF = 3
Duration 12-41 = 2
Independent float = E, - L? - D ? - ~
Late start schedule
=8-3-2=3
Independent float = 3

0 2 4 6 8 10
(c)

Figure 4-2 Network illustrating independent float.

float of 3 days. For an arbitrary activity (i - j), this can be written as:

The interpretation is that it has this float "independent" of any slippage of pre-
decessor activity 1-2, and any allowable start time of successor activity 4-5. In
contrast to this, activity 4-5 in Table 4-1, has no independent float and hence
its float of 2 days is not "independent" of any slippage of its predecessor activ-
ity 2-4.
Critical path computer packages always list path (or total) float for each activ-
ity, they occasionally list activity (or free) float, and t o the best of our knowl-
edge, they never listany other forms of float. That is not to say that they have
no utility, e.g., in tasks such as resource scheduling, or computerized bar chart-
ing. However, these other forms of float are currently judged t o be very margi-
nal, and are not utilized in this text.

USE OF SPECIAL SYMBOLS I N SCHEDULING COMPUTATIONS

Although there are a number of obvious advantages to having the network drawn
t o scale on a time base as shown in Figure 4-1,the disadvantages of doing this by
hand, notably the inflexibility to incorporating network changes, preclude the
general use of this procedure. It has been found best in practice not to attach
any special significance to the length of the network arrows, but rather to de-
note the various activity and event times of interest by numerical entries placed
directly on the network. As an aid to making the scheduling computations
which give these numerical entries, and to display them in an orderly fashion so
that they can be easily interpreted, the authors have developed a system incor-
porating special symbol^.^ These symbols provide spaces on arrows and in event
nodes for recording computed times. The spaces are located logically to bring
activity earliest finish (and latest allowable start) times close together and thereby
facilitate the computation of earliest expected (and latest allowable) event times.
These symbols are used throughout this text t o make it easier for the reader to
follow. In practice, however, the authors omit the enclosures for LSij, Fii,and
E h , shown in Figure 4-3a; these numbers are merely written above a single line
activity arrow in the appropriate position.
The event identification number is placed in the upper quadrant of the node,
as shown in Figure 4-3a. The earliest event occurrence time, Ei, which is equal
to the earliest start time (ESli) for activity i - j, is placed in the left hand quad-
rant of event i. The latest allowable event occurrence time, Li, which is equal to
the latest allowable finish time (LFii) of activity i - j , is placed in the right hand
quadrant of event j. The lower quadrant might be used later to note actual event
occurrence times if desired. For each activity, the earliest time the activity is ex-
pected to be finished, EF, is placed in the arrow head and the latest allowable
 Basic Scheduling Computations 83

Reading earliest expected and latest allowable activity start

and finish times and float from the special symbols

Total float

Pp9
Free float

i
?!$ L$ 6, ECi
0, F"
'i

Earliest Latest Earliest Latest

Stan times Finish rimes
for activity li - j ) for activity li - j )
Figure 4-3a Key t o use and interpretation of special activity and event symbols.

start time, LS, in the arrow tail. The estimated duration of the activity along
with a description of the activity is placed along the arrow staff. Path float is
placed in the bubble along the arrow staff. The lower portion of Figure 4-3a
points out how one reads the earliest and latest allowable start and finish times
for an activity. The detailed steps involved in carrying out the scheduling corn-
putations using these symbols is illustrated in Figures 4-3b and 4-4.

Begin with zero for the earllex srart time for the initial project evernand complte
, -
earlien finish timer for all succeeding sctlvitier For a typical activity, plme i t s Whereactivities merge, inrert in the l e f l quadrant
earliesrxan time imq. 23 days from project n a n l in the left quadrant of theevent Of the went SYmbOi rhe l a m * of the earlien finish
Flmbol. T k n add it%duration 171 t o theesriiest starttime to obmin ilr earlim timer written in the arrowhwd$of the merging
finish r i m 1301. Write 30 in rhe arrow head. mt~viTies.

Backward pass

Place the scheduled complerion time for the final went in the right quadrsnt of the y
P'OIB(I ler~u~rtal
Nenl spin bur olner wmm. Insert i.3e.d The atell allc*aole fine,, fvrv or r "bllm" fmm an n m t .
a ~ ~ 1nrr
N C W occ-rrence t me For a t)p cal rctrr I ) . r o n a L t onr al~ra!mor(I1 from ~ h r
r n t n in the r ght q~a6ranrof tlw event symbol the
lrrerr a,mn w o t ~~ n 1351
r 10 1 nta n the aten s lo*aole acts t i *an 1 n e ,281.
.ma1 ert of b t m a. ovablr act r i w rton limes
W'r In 28 n me a m w m

Figure 4-3b Steps in scheduling computations using special activity and event symbols.
84 l 1 Basic Topics

(TOTAL) PATH FLOAT

F.28- 23 = 5 = 35 - 30

A C T I V I T Y (FREE1 FLOAT

Figure 4-4 Steps in path float and activiw float computations using special activity and
event symbols.

ILLUSTRATIVE NETWORK EMPLOYING SPECIAL SYMBOLS

To illustrate the use of these symbols in making the scheduling computations, a

network containing eleven activities is shown in Figure 4-5, which contains the
complete forward pass computation. In Figure 4-6 the backward pass and float

Figure 4-5 Illustrative network employing the special activity and event symbols showing
f o r w r d pass computations only.
 Basic Scheduling Computations 85

computations have been added. With a little practice, the computations made
on this network can be completed in two to three minutes. With an allowance
for an independent check on these computations, the total time is still less than
that required to fill out the input forms required to run this problem on a com-
puter. This holds true for networks of any size. The computer is only economi-
cal when the network is to be updated periodically, or when a variety of other
computations is requested.
In Figure 4-6, the earliest and latest allowable activity start and fiiish times
are clearly displayed to aid in the making of resource allocation checks, the de-
termination of activity schedules, the conducting of time-cost trade-off studies,
etc. For example, consider activity 2-5. Its earliest start and finish times are
readily observed to be 6 and 7, respectively, and its latest allowable start and fi-
ish times are 11 and 12, respectively. The path float is F 2 , = LF,, - EF,, =
12 - 7 = 5 days, while the activity float is only AF,, = E, - EF2, = 11 - 7 = 4
days. This activity illustrates quite well the basic difference between path float
and activity float, the latter occurring only at the end of a slack path, i.e., at a
merge event. If the completion of activity 2-5 is delayed up t o 4 days, the
amount of its activity float, no other activity or event time in the network will
be affected. In particular, the earliest expected time for event 5 remains at 11.
If the completion of this activity is delayed by an amount exceeding its activity
float, but not exceeding its path float, then the earliest expected time for event
5 and the early start time for the following activity 5-8 will be increased. How-
ever, no critical path activities or events, such as event 8, will be affected. Fi-
nally, if the completion of activity 2-5 is delayed by an amount which exceeds

Figure 4-6 Illustrative network employing the special activity and event symbols showing
completed computations.
86 l 1 Basic Topics

its path float of 5 days, then the project completion time, i.e., the earliest ex-
pected time for event 8, will be increased by a like amount.
According to the previous definition of the critical path, it is made up of ac-
tivities 0-3-7-8 in the illustrative network shown in Figure 4-6. It has the least
amount of path float, which is zero in this case, because the convention of let-
ting L8 = E8 = 15 was followed. This is also the longest path through the net-
work. In addition to determining the critical path through the network, we can
identify various subcritical paths which have varying degrees of path float and
hence depart from criticality by varying amounts. These subcritical paths can be
found in the following way, which is suggestive of how a computer would handle
this problem.

1. Sort the activities in the network by their path float, placing those activi-
ties with a common path float in the same group.
2. Order the activities within a group by early start time.
3. Order the groups according to the magnitude of their path float, small val-
ues first.
The first group comprises the critical path(s) and subsequent groups
comprise subcritical paths of decreasing criticality.

Application of the above procedure to the network in Figure 4-6 gives the re-
sults shown in Table 4-2 below.

Table 4-2. Listing of Critical and Subcritical Paths by Degree of

Criticality for the Network in Figure 4-6

Earliest Latest

Start Finish Start Finish Tolal

Activity Time Time Time Time Float Criticality

03 0 2 0 2 0
3-7 2 10 2 10 0 critical path
7-8 10 15 10 15 0
3-4 2 7 3 8 1
4-5 7 11 8 12 1 a "near critical" path
5-8 11 14 12 15 1
0-1 0 2 5 7 5
1-2 2 6 7 11 5 third most critical path
2-5 6 7 11 12 5

06 0 1 6 7
path having most float
6-7 1 4 7 10
CRITICAL PATH FROM FORWARD PASS ONLY

The above procedure for locating the critical path(s) is based on a knowledge of
path float, which requires the backward pass for computation. While this pro-
cedure is necessary to find the slack along subcritical paths, the critical path(@
can be determined from the results of the forward pass only. This is quite useful
in the early stages of planning and scheduling a project, when it is desired to de-
termine the expected project duration, and to determine the critical path activi-
ties with a minimum of computation. The following steps which make up this
procedure are based on the assumption that the forward pass computations have
been completed, and the resulting EF's and E's have been recorded on the
network.

1. Start with the project final event, which is critical by definition, and pro-
ceed backwards through the network.
2. Whenever a merge event is encountered, the critical path(s) follows the ac-
tivity(~)for which EF = E.

To illustrate this procedure, let us trace the critical path of the network
shown in Figure 4-5, on which only the forward pass computations have been
made. First we start at event 8 for which E = 15. The critical path is then along
activity 7-8 since E F = E = 15 for this activity, while EF is only 14 along the
other path, activity 5-8. In this manner, the critical path can be traced next to
event 3, and hence to the initial network event 0.

VARIATIONS OF THE BASIC SCHEDULING COMPUTATIONS

The restrictive assumptions underlying the above basic scheduling computations

included the following:

1. The network contained only one initial event, i.e., one event with no pre-
decessor activities.
2. The network contained only one terminal event, i.e., one event with no
successor activities.
3. The earliest expected time, E, for the initial event was zero.
4. There were no scheduled or directed dates for events other than the net-
work terminal event.

These assumptions were made because they simplify the computational pro-
cedure and its subsequent interpretation, and at the same time they do not se-
riously restrict the usefulness of the procedure. However, occasions will be
pointed out where it would be of some value to relax these assumptions; hence,
88 I I Basic Topics

the required modifications in the computational procedures will be taken up

here.
It may happen, for example, that the network under study is only a portion
of a larger project, or perhaps one of several in a multi-project operation. In this
case, it may be that the earliest time for the initial network event does not occur
at time zero, but rather at some arbitrary number of time units other than zero.
In such a case, this specified earliest occurrence time, E, for the initial event is
merely used in place of zero, and the forward pass computations are made in the
conventional manner.
Similarly, one may wish to specify or direct the latest allowable time for some
intermediate network (milestone) event to be a time that is arbitrarily specified
as the scheduled or allowed time for the event in question. In this case one uses
a scheduled time, Ts, in determining the L value for this event, as described in
the example below. The important points in this discussion are summarized
below.

Conventions:

A scheduled time, Ts, for an initial project event is interpreted as its earliest
expected time, i.e., Ts = E for initial project events.
A scheduled time, Ts,for an intermediate (or terminal) project event is in-
terpreted as its latest allowable occurrence time.

ILLUSTRATIVE NETWORK WITH MULTIPLE INITIAL

AND TERMINAL EVENTS

To illustrate how projects with multiple initial and terminal events and sched-
uled event times are handled, consider the network shown in Figure 4-7. The
"main" project in this network has events numbered 101 through 109. This
project, which starts with event 101, produces an "end" objective signified by
event 109. However, a by-product of this project is that it furnishes an output
for a second project whose events are numbered in the 200 series. For example,
the objective of the main project might be to develop a new rocket engine, the
completion of which is denoted by event 109. Event 210 mtght signify the de-
livery of a key component from a second project that has other end objectives.
Activity 108-211 might be the preparation and delivery of a report on the test-
ing of this component which took place in the main project. The scheduled
occurrence time of event 21 1 is thus quite important t o the second project, and
is independent of the main project.
Two events from the second project are pertinent to the main project network
shown in Figure 4-7. First, event 210, which initiates an activity preceding event
106, is scheduled to occur 8 units of time after the start of the first project.
Hence, if one sets E = 0 for event 101, then E = 8 for event 210, since the latter
 Basic Scheduling Computations 89

Figure 4-7 Network with multiple initial and end events and scheduled event times.

is one of the initial events in the network. Event 211 of the second project is
scheduled to occur 14 units of time after the start of the first project. Hence,
since event 21 1 is a terminal event in this network, one sets Ts = L = 14 for
event 211. Finally, the main project is scheduled to be completed in 20 time
units, thus Ts = L = 20 for event 109. Thus, the network computations are
started with the following times specified.
Event 101: E=O
Event 109: L =Ts=20
Event 210: E = Ts = 8
Event211: L = T s = 14
Since early start times are now specified for both of the initial events in this
network, the forward pass computations can be carried out in the conventional
manner. These computations indicate that the early finish time for the final
event of the main project is 18, i.e., E = 18 for event 109. The early finish time
for the secondary project is 15, i.e., E = 15 for event 211. Thus, the schedule
can be met on the main project but not on the secondary project, unless the lat-
ter is expedited in some way.
Since all terminal events have scheduled completion times, the backward pass
computations are initiated by setting L = 20 for event 109, and L = 14 for event
211. The critical and subcritical paths through the network can then be deter-
mined as given in Table 4-3.
The critical (least float) path through the network has a path float of - 1. One
90 I I Basic Topica

Table 4-3. Critical and Subcritical Paths for the Network

in Figure 4-7

Activities Path Float, F Activities Path Float, F

210-106 -1 103-104 1
106-108 -1 104-109 1
108-211 -1
106-109 2
101-102 0
102-103 0 102-107 3
1 W-105 0 107-108 3

thus expects t o be one unit of time late in meeting the scheduled time of 14 for
event 211, the termination of the secondary project. With regard to the main
project, one notes that all paths leading to the terminal event 109 have path
float of at least one time unit.
Multiproject networks can be quite useful in analyzing the effects of one proj-
ect on another, and thus offer a means of settling disputes which frequently arise
in such a situation. For example, it is clear from Figure 4-7 that one way of
alleviating the negative slack situation on the critical path would be to move up
the schedule for event 210 so that it occurs on or before the late start time of 7
for activity 210-106. In this way the secondary project could help in solving its
own scheduling problem. If this could not be done, then the only remaining
remedies would be to reduce the duration of either activity 210-106 or 108-211
by one or more time units. It is also clear from Figure 4-7 that the current
schedule for event 210, i.e., Ts = 8, does not produce the most constraining path
to event 109, and hence does not fuc the earliest completion time of the main
project.
Another method of handling multiple initial and end events is taken up in
exercise 4 at the end of this chapter. This procedure is illustrated in Figure 3-16
where trt-o essentially separate projects are tied together by activity 40-80, which
has a time estimate of 40 days. In essence, this activity constrains the comple-
tion of one project to precede the other by 40 days. Thus, a scheduled date
placed on event 80 in one project can force a scheduled date, or latest allowable
time, 40 days earlier on event 40 of the second project. This procedure elimi-
nates the second terminal event.

NETWORKS WITH SCHEDULED OR DIRECTED TIMES

ON INTERMEDIATE EVENTS

Scheduled or directed times for intermediate network events are handled in a

manner similar to the treatment for terminal events as discussed above. In this
 Basic Scheduling Computations

case, however, there will be two candidates for the latest allowable time f o r t
event in question. The choice is usually governed by the foflowingconven
91
however, other conventions may be adopted in certain project management
puter packages.
The latest allowable time for an intermediate network event on which a sche
uled time, T s , is imposed, is taken as the earlier (smaller) of the scheduled
Ts, and the latest allowable time, L , computed in the backward pass.
For example, suppose activity 103-105 in Figure 4-7, which involves eart
moving, was scheduled to be completed by time 7 to insure completion prio
the ground freezing. In this case, L for event 105 would be taken as 7, i.e.,
smaller of the scheduled time of 7 and the regular backward pass time of 9.
introduction of this scheduled time only affects the path float for activities 10
102-103-105 by reducing it from 0 t o -2, a change which is quite important
planning and scheduling these activities. Since this now becomes the path wi
the least float, it forms the new project critical path. While it starts at the initi
project event, it terminates at the intermediate event 105 on which the sched
uled time was imposed, without going through the entire project. This is a cha
acteristic result of introducing scheduled or directed times on intennediat
events; the computations clearly show up the most constraining scheduled datek
in the project.
I
NETWORK TIME-STATUS UPDATING PROCEDURE

Updating a network to reflect current status is similar to the problem

1
above in that a project underway is equivalent to a project with
events. After a project has begun, varying portions of each path from the initi
project event to the end event will have been completed.
diagram network logic, all activities preceeding the current status on a particul
path must be completed, and no activities succeeding the
should have started. By establishing the current status on
ress information, the forward pass scheduling computations can then be made ir
routine fashion. No change in the backward pass computation should be neces,
sary, unless changes are to be incorporated in the logic or time estimates of the
uncompleted portion of the network.
To illustrate this updating procedure, consider the network presented in Fig.
ure 4-6, which indicates an expected project duration of 15 days. Suppose the
project started on Monday and we have just received the Friday evening week-
end progress report. Thus, we have just completed the fifth work day on this
project, and the progress is as reported in Table 4-4.
The actual activity start and finish times given in Table 4-4 have been writter
above the arrow tails and heads, respectively, in Figure 4-8. Events that have al-
ready occurred have been cross hatched, and activities that are in progress have
92 l l Basic Topics

Table 4-4. Status of Project Activities at the End of the Fifth Working Day

Activity Started Finished

NOTE: all times given are at the end ol the stated working day.

been so noted by a flag marked 5 to denote that the time of the update is the
end of the fifth working day. Only one of the four paths being worked presents
a problem, i.e., activity 3-4. At report time, event 3 has occurred, but activity
3-4 has evidently not started. To avoid this problem, it would be desirable to in-
clude in progress reports the intended start time of all activities whose predeces-
sor activities have been completed. If this information is not given, then some
assumption must be made to complete the update calculations. The usual as-
sumption, which is the one adopted in Figure 4-8, is that the activity will start
on the next working day, i.e., at the end of the fifth working day.
Having an actual, or assumed, start time for the "lead" activities on each path
in the network, the forward pass calculations are then carried out in the usual
manner. The original times are crossed out, with the new updated times written

P
OeMfeS IlatYV 0f
acflvity at the end
of the 5th work day

Figure 4.8 Illustrative network showing time status of project.

 Basic Scheduling Computations

nearby. These calculations indicate that the critical path has shifted t o act
3-4-5-8, with a slack of minus two days. Assuming we were scheduled t
I
3

plete the project in 15 days, the current status indicates we are now two d
hind schedule, and appropriate corrective steps are in order.
An alternative updating procedure is as follows; it has the advantage
can be carried out by a computer program that has no special updating
ity. First, add a dummy activity preceding the initial project event, and ass
to it a duration value equal to "time now," that is, the number of elapsed wo
ing time units from the beginning of the project to the time of the pr
date. Second, change the duration times of activities that are complete
tially completed, to the number of time units of work remaining. N
the forward pass calculations in the conventional manner. It will pr
same results obtained by the method illustrated in Figure 4-8 above.

ACTIVITY-ON-NODE SCHEDULING COMPUTATIONS

In Chapter 2, the basic node scheme of networking was described. This is

I
complete reverse of the arrow scheme in which
and the arrows are merely connectors. To illustrate the
dure for a node diagram, consider the network plan for
ect presented in node diagram form in Figure 2-21.
gram was given in Figure 2-18, and a time-scaled
basic forward and backward pass scheduling
given in Figure 4-9.

Key

Figure 4-9 A manual computational method for the activity-on-node scheme.

94 I I Bssic Topics

The node scheme, first promoted by J. W. ~ o n d a h in

l ~ 1958, lends itself to
manual computation as well as the arrow scheme. One symbolic method for
manual computation of node networks is illustrated in Figure 4-9. Using the key
to the symbol notation, the reader can easily see how the computation is made.
For example, the EF for node 3 is obtained by adding its times for ES and D.
Thus, EF = 4 + 2 = 6. At merge points, such as node 7, the ES is selected as the
largest of the preceding EF times. The backward pass is made similarly.
In comparison with the symbols used in manual computation of arrow net-
works, the node symbols are somewhat more articulate. All of the numbers
associated with an activity are incorporated in the one node symbol for the ac-
tivity, whereas the arrow symbols contain each activity's data in the predecessor
and successor nodes. as well as on the arrow itself.

PRECEDENCE DIAGRAMMING

An important extension to the original activity-on-node concept appeared

around 1964 in the Users Manual for an IBM 1440 computer program.' One of
the principal authors of the technique was J. David Craig, who referred to the
extended node scheme as "precedence diagramming." The computation and in-
terpretation of earlyllate startlfinish times for project activities for this scheme
are considerably more complex than those shown above for the basic finish-to-
start constraint logic of arrow or node diagrams. For the latter, the computation
and interpretation of these times was both simple and unique. This is an impor-
tant advantage of the PERT/CPM systems. Unfortunately, this does not usually
hold for precedence diagrams; a number of complications can arise, as will sub-
sequently be shown.
The basic computational approach to be used in this text is to adopt a pro-
cedure that will lead to activity earlyllate start/fiiish times for a precedence
diagram network that are identical to those that would be obtained for the
equivalent arrow diagram and the conventional forward and backward pass com-
putations. This approach was developed by Keith C. Crandall, Professor of Civil
Engineering, University of California, Berkeley, and published in 1973' A hand
computation version, where activity splitting is allowed, was also supplied by
Crandall in private communication in May 1981; it is given in Appendix 4-2 of
this chapter.

Definitions:

The computational procedure to be given here is based on an extension of the

PERT/CPM network logic from a single finish-to-start type of dependency to in-
clude three other types. These other dependency relationships were introduced
in Chapter 2, and are repeated in Figure 4-10 below, in slightly different form.
 Basic Scheduling Computations
I
5

ss,,
Start-to-start constraint Finish-to-finishconstraint
I
Finish-toltart constraint

Figure 4-10
SF..I, =SF.'I +SF."
8,

Startto-finish constraint

Precedence diagramming constraints w i t h leadhag times.

I .

The nomenclature and assumptions to be used are as follows:

SSij denotes a start-to-start constraint, and is equal to the minimum num

I
of time units that must be complete on the preceding activity (i) prio
the start of the successor ( j ) .
FFil denotes a finish-to-finish constraint, and is equal to the min
ber of time units that must remain to be completed on the successor
after the completion of the predecessor (13.
FSit denotes a finish-to-start constraint, and is equal to the minim
of time units that must transpire from the completion of the pr
(9 prior to the start of the successor (0. (Note: This is the
constraint used in PERTICPM, with FSIj= 0).
SFij denotes a start-to-finish constraint, and is equal to the minimum nu
of time units that must transpire from the start of the predecessor
the completion of the successor (j).
ZZfj denotes a frequently used combination of two constraints,
to-start and a finish-to-finish relationship. It is written w
time units first, followed by the FFij time units.

The above constraint logic is shown in Figure 4-10. It will be applied in

next section to illustrate the powerful features of the extended logic of
96 l 1 Basic Topics

dence diagramming. It will also point up an important anomaly that can occur,
and needs an explanation.

Precedence Diagram Anomalies

Consider a construction subcontract consisting of Fmming walls, placing E1ecm.-

cal conduits, and Finishing walls, with the duration of each task estimated to be
10 days, using standard size crews. If the plan is to perform each of these tasks
sequentially, the equivalent arrow diagram in Figure 4-1 l a shows that a project
duration of 30 days will result.
To reduce this time, these tasks could be carried out concurrently with a con-
venient lag of say 2 days between the start and finish of each activity. This plan
is shown in Figure 4-11 b in precedence diagram notation. The equivalent arrow
diagram shown in Figure 4-1 1c indicates a 14 day project schedule. One impor-
tant advantage of Figure 4-1 1b over 4-1 1c is that each trade is represented by a
single activity instead of 2 or 3 subactivities. Also note how the SS= 2 and
FF= 2 lags of Figure 4-1 1b are built into the equivalent arrow diagram in Fig-
ure 4-1 lc. For example, the first two days of the electrical task in Figure 4-1 1c
must be separated from the remainder of this task to show that 2 days of electri-
cal work must be completed prior to the start of the finishing task. Similarly,
the last 2 days of electrical work must be separated from the remainder of this
task to show that framing must finish 2 days before electrical is finished. Thus,
the 10 day electrical task must be broken up into 3 subactivities of 2,6, and 2
days duration, respectively.
So far, precedence diagramming is easy to follow and is parsimonious with ac-
tivities. But let us see what happens if the duration of the 3 tasks in this project
are unbalanced by changing from 10, 10, 10 to 10,5, and 15 days, respectively.
These changes are incorporated in Figures 4-1 1d and e, along with appropriate
new lag times. Note that SS = 2 was chosen between framing and electrical to
insure that a full days work is ready for electrical before this task is allowed to
start. Similarly, FF= 3 was chosen between electrical and finishing because the
last day of electrical work will require 3 days of finishing work to complete the
project. The other lags of 1 day each were chosen as minimal or convenience
values needed in each case. These lags define the activity breakdown shown in
Figure 4-1 l e where we see the critical path is the start of framing (1-2), then the
start of electrical (4-S), and, finally, the totality of finishing (8-9-10). This is
also shown in the precedence diagram, Figure 4-1 Id, where ES = LS = 0 for the
start of framing, ES = LS = 2 for the stmt of electrical, and, finally, ES = LS = 3
and EF=LF = 18 for the totality of finishing. Since the precedence diagram
shows each of these tasks in their totality, EF # LF even though ES = LS for
the framing and electrical tasks. For framing in Figure 4-1 Id, LF - EF= 14 -
10 = 4 days of float, which corresponds to the 4 days of float depicted by activ-
ity 7-9 in Figure 4-1le. Similarly, for electrical in Figure 4-1 Id, LF - EF = 15 -
 Critical path = 1 - 2 - 3 - 4
Framing
I
Electrical
I
Finishing
Duration = 10 + 10 + 10 = 30 days
I
-3 4

0 4 8 12 16 20 24 28 32 Time (days)
(a1 ~equentiaiarrowdiagram I

Time (days)
Figure 4-11 Arrow and precedence networks t o illustrate splitting vs. n o splitting; activi.
ties are shown a t their early start times.
98 l l Basic Topics

11 = 4 day?, of float, which is also depicted by activity 7-9 in Figure 4-1 le. The
middle electrical activity (5-6) in Figure 4-1 l e a p p e m to have an additional path
float of 4 days, or a total of 8 days. This attribute is not shown at all in Figure
4-1 1d because it depicts only the beginning and end points of each activity, but
not intermediate subactivities such as 5-6. Closer examination will show, how-
ever, that any delay in the start of activity 5-6 exceeding 3 days would cause the
finishing crew to run out of work, and hence the critical path would be delayed.
This problem is shared by both arrow and precedence diagrams, and the user
should understand this. It does not, however, present a real problem in the ap-
plications since the job foreman generally has no difficulty in the day-to-day
management of this type of interrelationship among concurrent activities. It is
generally felt that it is not worthwhile t o further complicate the networking and
the computational scheme to show all interdependencies among activity seg-
ments, since these tasks can be routinely managed in the field.
A very important difference between Figures 4-1 1c and e, other than the 4
day difference in the project durations, lies in the electrical task, which is repre-
sented by 3 subactivities in both diagrams. In Figure 4-1 1c these 3 subactivities
are expected to be conducted without interruption. However, in Figure 4-1l e
this is not possible. Here, the last day of the electrical task (6-7) must follow a
4 day interruption because of the combination effect of successor constraint SS1
depicted by activity 5-8, and predecessor constraint FF1 depicted by activity
3-6. These constraints require events 5 and 6 to be separated 7 days in time,
while the intewening electrical activity requires only 3 days to perform. This
forced interruption will henceforth be referred to as splitting of the electrical
task.
If necessary, splitting can be avoided in several ways. First, the duration of
the electrical task could be increased from 5 to 9 days. But this is frequently
not desirable in projects such as maintenance or construction because it would
decrease productivity. The second way to avoid splitting would be to delay the
start of the electrical task for 4 days, as shown in Figure 4-1 lg, where it is as-
sumed that activity splitting is not allowed. At fist, it may seem that there is no
difference between these two alternatives, but this is not so. Reflection on Fig-
ure 4-1 lg shows that delaying the start of the electrical task to avoid splitting
will delay the start of the finish work, and hence the completion of the project is
delayed by 4 days. But increasing the duration of the electrical task will not
have this effect. Actually, we have described an anomalous situation where an
increase of 4 days in the duration of an activity on the critical path (starting 4
days earlier and thus running 4 days longer), will decrease the duration of the
project by 4 days, from 22 to 18. If you are used to dealing with basic arrow
diagram logic (FS= 0 logic only), this anomaly will take some getting used to.
It results from the fact that the critical path in Figure 4-1 1g goes "backwards"
through activity 56,and thus subtracts from the total duration of this path. As
 Barie Schaduling Computations

a result, the project duration shifts in the reverse direction of a shift in the
b
I

tion of such an activity. That is, the project duration decreases when the
ity duration increases, and increases when the activity duration decreases.
anomalous situation occurs whenever the critical path enters the
an activity through a finish type of constraint
through the activity, and leaves through a
The precedence diagram in Figure 4-1
cal, since ES = LS and E F = L F for each task. While it
task has float in Figure 4-1 lg, this is not true since splitting is not allowed.~b -
splitting is a constraint not explicitly incorporated in the arrow diagram logic. 1
Critical Path Characteristics

iI
Wiests describes the anomalous behavior of activity 5-6 in Figure 4-1 1g pict
esquely by stating that this activity is reverse critical. Similarly, in Figure 4-1 d
and e both framing and electrical are called neutral critical. They are critical b -
cause their LS = ES, but they are called neutral because their LF >EF, and t e
project duration is independent of the task duration. A task is neutral critic
when a pair of start time constraints result in the critical path entering and exi -
ing from the starting point of the task, or apair of finish time constraints entkr
and exit from the finish point of a task. These situations could also be referre
to as start or finish critical. In Figure 4-1 Id and e, the framing and electric
tasks are both start critical, while finishing is normal critical. That is, a shift i
the duration of the finishing task will have a normal effect on the project dur -
tion, causing it to shift in the same direction. Wiest suggests that precedenc(e
diagram computer outputs would be more useful if they identified the way i/l
1
t

I
which tasks are critical. The author suggests that the following nomenclature
considered for this purpose:

NC denotes an activity that isNormal Critical; the project duration shifts i

the same direction as the shift in the duration of a NC-activity.
RC denotes an activity that isReverse Critical; the project duration shifts i
the reverse direction as the shift in the duration of a RC-activity.
BC denotes an activity that is Bi-Critical; the project duration increases as
result of any shift in the duration of a BC-activity. (See exercise 13
the end of this chapter for an example.)
SC denotes an activity that is Start Critical; the project duration shifts i
the direction of the shift in the start time of a SC-activity, but is neutr
(unaffected) by a shift in the overall duration of the activity.
FC denotes an activity that is Finish Critical; the project duration shifts i
the direction of the shift in the finish t i e of a FC-activity, but is n
tral (unaffected) by a shift in the overall duration of the activity.
100 l l Basic Topia

MNC denotes an activity whose Mid portion is Normal Critical.

MRC denotes an activity whose Mid portion is Reverse Critical.
MBC denotes an activity whoseMid portion is Bi Critical.

To conclude this discussion, it should be noted that the critical path always
starts with a job (or a job start), it ends with a job (or a job finish), and in be-
tween it consists of an alternating sequence of jobs and precedence arrows. Al-
though the critical path may pass through a job in any one of the many ways
listed above, it always moves fonvard through precedence constraint arrows.
Hence, any increase (decrease) in the lead-lag times associated with SS, SF, FF,
or FS constraints on the critical path, will always result in a corresponding in-
crease (decrease) in the project duration.
Following the suggestion of stating the nature of the criticality of activities on
the critical path, for Figure 4-1 Id this would consist of the following alternating
activities and precedence constraints: Framing (Start CriticalSC); SS2; electri-
cal (Start Critical-SC); SSl; finishing (Normal Critical-NC). Similarly, for Fig-
ure 4-1 1f it would be: Framing (NC); FFl; electrical (RC); SSl; finishing (NC).
It should be noted here that electrical is labeled reverse critical (RC), which puts
the manager on notice that any shift in the duration of this activity will shift the
duration of the project in the reverse direction. As stated above, it is reverse
critical because its predecessor constraint is a finish type (FFI), and its successor
constraint is a start type (SSl).

PRECEDENCE DIAGRAMMING COMPUTATIONAL PROCEDURES

Obviously the forward and backward pass computational problem becomes more
complex with precedence diagramming, and it calls for establishment of some-
what arbitrary ground rules that were unnecessary with the unique nature of
basic arrow diagram logic. In the computational procedures to follow, we will
assume that the specified activity durations are fixed, e.g., because of the pro-
ductivity argument cited above. This assumption can be relaxed, of course, by
varying the activity durations of interest, and repeating the calculations. Regard-
ing task splitting, three basic cases will be treated.

Case I : Activity splitting is not allowed on any activities.

C8e 2: Activity splitting is allowed on all activities.
Case3: Combination of 1 and 2; activity splitting is permitted only on desig-
nated activities.

Figures 4-1 l g and e represent Cases 1 and 2, respectively. The effect of not
allowing splitting (of the electrical task) is a 4 day increase in the project dura-
 Basic Scheduling Computations 101

tion. Here, the choice must be made between the (extra) cost of splitting {he
electrical task, and the cost of a 4 day increase in project duration. Case is 4
provided to allow the project manager to take the possible time (project duratidn)
advantage concomitant with splitting on those activities where it can be tojer-
ated, and to avoid splitting on those activities where it cannot be accommodated.
The computational procedure for Case 1 is reasonably simple and will be e-
scribed below. The procedure for Case 2 is considerably more complex; i is 1
given in Appendix 4-2. The computational procedure for Case 3 merely amou ts
1
to the application of the Case 1 or the Case 2 procedure to each activity in tu n,
depending on whether the activity is designated as one where splitting is ot
allowed, or is allowed, respectively. 1
Computational Assumptions

The computational procedure for Case I-No Splitting Allowed, is analogous to

~
the arrow diagram procedure described above. In making the forward pass oal-
culations, one must consider all constraints leadine into the activitv (il in oups-
- -,and FSii)
tion, i.e., the start time constraints .(SSii -,- as weNas the finish time cqn-
straints (SFli and FFii). For each constraint, the early start time for activity :J?
is computed, and the maximum (latest) of these times then becomes the ea.ly
start time (ESj) for activitv (i).Because some project activities may only hqve
finish time constraints, it would be possible for the above procedure tolead to a
negative ESi time, or a time earlier than the specified project start time. For rx-
ample, referring to Figure 4-12, we see that activity D has no start time ccn-
straint. If the duration of activity D was 22 (instead of 12), then its early str.rt
time would be E F - D = ES, or 19 - 22 = -3 (instead of 7). This would be an
erroneous negative value. To prevent the occurrence of this error, an additior al
time, called the INITIAL TIME, is introduced. It is usually set equal to zero, or
else to an arbitrarily specified (nonzero) project scheduled start time, and it
overrides the start times computed above if they are all negative, or less (earlier)
than the specified project start time.
The backward pass computations follow a similar procedure to find the late
finish times for each activity, working backwards along each constraint leaviig
the activity (9 in question. In this case, an additional time, called TERMIN4L
TIME. is reauired to urevent the occurrence of a late finish time
.
the project duration, or the scheduled project completion

The computational procedure given

102 l 1 Basic Topics

0 12 FF5 7 19 19 28 FF7

Critical

FS 7
Key:
- 28 39 35 42
ES EF

constraint
'22 3, 5 55 3 FF 5

durations
LS LF
Figure 4-12 Example network with forward and backward pass times shown, and the nota-
tion of the critical path-no splitting allowed.

activity will always be found below it in the ordered list. The two step computa-
tional procedure is then applied to each activity working the list from the top
down. When the computations are performed by hand on a network, this order-
ing is accomplished automatically by working one path after another, each time
going as far as possible. Again, this is the same procedure required t o process an
arrow diagram.

Forward Pass Computations-No Splitting Allowed

The following two steps are applied to each project activity, in topological se-
quence. The term called INITIAL TIME is set equal to zero, or to an arbitrarily
specified project scheduled start time.
 Basic Scheduling Computations 1q3

I
STEP 1 . Compute ESi, the early start time of the activity (19 in question. t
is the maximum (latest) of the set of start times which includes t e
INITIAL TIME, and one start time computed from each constra' t
going to the activity (13 from predecessor activities indexed by (i).

INITIAL TIME

1 ~+sSQ~- D~J
STEP 2. EFi = ESi + Dj

Backward Pass Computations-No Splitting Allowed

The following two steps are applied to each project activity in the reverse ord
~
of the forward pass computations. The term called TERMINAL TIME is s
equal to the project duration, or to an arbitrarily specified project
completion time.

STEP 1. Compute LFi the late finish time of the activity (i) in question. It
the minimum (earliest) of the set of finish times which includes
TERMINAL TIME, and one finish time computed from each
straint going from activity (i), to successor activities indexed by

LS' - FS$

all i
LSi - SSrj + Di
LFj - SFij + D l

STEP 2. LSI =LFi - Di 1

Example Problem

To illustrate the application of the above algorithm, a small network consistinb

I
of 9 activities with a variety of constraints, is shown in Figure 4-12. The
ward pass calculations are as follows, based on the assumption that the projec
starts at time zero, i.e., INITIAL TIME = 0.
104 1 I Basic Topia

Activity A
ESA = (INITIAL TIME = 0) = 0

EF,=ESA+DA=0+12=12

Activity B

i
INITIAL TIME = 0

ESB=MAX E S A + S S A B = 0 + 3 = 3
A
EFA +FFAB
-DB=12+5- 10x7
EFB=ESB+ D B = 7 +l o = 17

Activity D
INITIAL TIME = 0
ESD = MAX 1.7
ESB+SFBD-DD=7+(2+10)12=7

Activity C

i
INITIAL TIME = 0

ESc=MAX E S A + S S A c = 0 + 3 = 3
A, D
ESD +SFDc - D c = 7 + ( 7 + 14)-
EFc=ESc + D c = 10+ 1 8 ~ 2 8
etc.

The backward pass calculations are as follows, wherein the TERMINAL TIME
is set equal to the project duration, determined from the forward pass calcula-
tions to be 42, i.e., the E F time for the last critical path activity G.

Activity G
LFG = {TERMINAL T I ME = 42) = 42
LSG=LFG-DG=42- 10=32

Activity I
LFI = {TERMINAL TIME = 42) = 42

LSI=LFI - D 1 = 4 2 - 7 = 3 5
 Basic Scheduling Computations
'P"
Activity H
i

Activity F

LFF=MIN
G ,I

L S F = L F F - D F = 3 5 -1 1 = 2 4
etc. I
From the computational results shown in Figure 4-12, the critical path co -

i
sists of activities A-B-D-E-F-G. The nature of the criticality of each activity is
indicated at the top of Figure 4-12, along with the critical constraints betwe n
each pair of activities. Activities A, E, and G are normal critical, activities B a d
F are rwerse critical (noted by the reverse direction cross-hatching), and activi y
D is only finish time critical. The duration of the critical path, 42, is also note ,
with the net contributions of the activity durations being (12 - 10 + 0 + 9 -
11 + 10) = 10 and the contributions of the constraints being (5 + 12 + 0 + 7 +
8) = 32, for a total of 42 time units. The earlyllate startlfinish times for each c-
tivity have the conventional interpretations. For example, for the critical acthr-

1
ity E, both the early andlate startlfinish times are 19 and 28; the activity has o
slack. But for activity H, the early startlfinish times are 24 and 35, while t e
late startlfinish times are 28 and 39. In this case, the activity has 4 units of -
tivity slack or free slack, because the completion of activity H can be delayed p
to 4 units without affecting the slack on its successor activity I.
To illustrate the use of computers for these computations, the complet d
basic schedding computations for this network are shown in Figure 4-13, usi g
a program written by K. C. Crandall. The upper portion of this figure giv s
times in elapsed working days, while the lower portion is given in calendar dat .
The latter assumes that start times denote the beginning of the day, and fin' h
times the end of the day. For example, activity A has a duration of 12
corresponding to an early start date of 3 August and an early finish date
August. This computer run assumes, of course, a 7 day work week. The
information is very conveniently displayed in bar-chart form in Figure
The upper portion of this figure gives the early-start activity schedule,
lower portion gives the late-start schedule. A study of this figure will
 OWNER: HODER TEXT - THIRD E D I T I O N COtJTRACTOR: I nil AN E X C E L L E N T CONTRACTOR
-
\

COtlPUTER S O L U T I O N SHOUING T H E .NON S P L I T T I N G ' O P T I O N C ? L C U L b T I O N UPDATE NET HOD 1

FOR THE T H I R D E D I T I O N OF THE MODER T E X T ON 'CPM L PERT r CALENDAR 3AUG81 16:55:11 RUN NO. 1
TEST CALCULATED OM PACKAGE B Y K C CRANDALL MODIFICATION 3hUG81 17:04138 ii'
CALCULATION 3AUG81 17:17:11 5
"
h C T l V I T Y SCHEDULE L I S T I N G - PAGE 1 ;
ACTIVITY START FINISH FLOAT
LhBL DESCRIPTION DURATION EhRLY LATE EARLY LATE TOTAL F R E E

THIS ACTIVITY
THIS ACTIVITY
THIS ACTIVITY
* CCCC
DDDD
EEEE
----
THIS
THIS
ACTIVITY
AC
- T-I V -
ITY
S FFFF THIS ACTIVITY
I GGGG THIS ACTIVITY
HHHH THIS ACTIVITY
1111 THIS ACTIVITY

A C T I V I T Y SCHEDULE L I S T I N G - PAGE 1

ACTIVITY START FINISH FLOAT

LnBL DESCRIPTION DUWATIDN EARLY LATE EARLY LATE TOTAL F R E E

t a m n T H I S I S A C T I V I T Y .Am 12 3 AUQ P I 3 AUG 81 14 AUG P i 1 4 AUG P1

t ABBB T H -I S- I- S
- A -C T-I V -I T Y 'B' 10 10 AUG PI 10 AUG 81 19 AUG 81 1 9 AUG P1
CCCC T H I S I S ACTIVITY 'C' IS 13 AUG 81 15 AUG e l 30 AUG si I SEP 81
t DDDD T H I S I S ACTIVITY 'Dm 12 10 AUG 81 10 AUG 81 21 AUG 81 21 W G 81
EEEE T H I S I S ACTIVITY .Em P 22
-- AUG- PI
- 2- -2 nuG e l 30 AUG PI 30 AUG 81
f FFFF THIS I S ACTIVITY 'Fm 11 27 AUG B I 27 AUG el 6 SEF 81 6 SEP 81
f GGGG T H I S I S ACTIVITY '6' 10 4 SEP 8 1 4 SEP 81 1 3 SEP 81 1 3 SEP el
HHHH T H I S I S ACTIVITY .Hm 11 27 AUG e l 31 AUG e l 6 SEP el 10 SEP 81
1111 T H I S I S ACTIVITY 'Im 7 4 SEP PI 7 SEP e l 10 SEP el 13 SEP el

Figure4-13 Time and date computer outputs for the basic scheduling computations on the network s h o w n in Figure 4-12--no splitting
allowed.
 OWNER: MODER TEXT - THIRD E D I T I O N CONTRACTOR: I A n AN EXCELLENT CONTRACTOR

COMPUTER SOLUTION SHOWING THE 'NON S P L I T T I N G ' OPTION CALCULATION

FOR THE T H I R D E D I T I O N OF T l l E HOVER TEXT ON .CPM I PERT..
". ...,,
CALENDAR
L

3 A U 0 8 1 16:55:11
,.-,
RUN NO.
,.-- .
I
TEST CALCULATED ON PACKAGE BY K C CRANDALL tlODIFICATION 3AUG81 17:04:38
CALCULATION 3 A U 0 8 1 17:17:11

x * I CPM E A R L Y START ~CTIVITY BARCHART * II PAGE: I- I

SYMBOLS USED ARE: <*'. C R I T I C A L PATH i <O> WORK DAY i ':-* TOTAL FLOAT i <.> HOLIDAY- UEEKEND i r-> UPDATE DATE

AUG 8 1 SEP 8 1

L A X DESCRIPTION CALNDR DAYS -> 10 20 30 40 50 60 70 80 90

* M A THS
I I S A C T I V I T Y .A' I***********
BBBB T H I S I S A C T I V I T Y '8. ******tat*
cccc THI S I
S ACTIVITY -c. ooooooo~~~~~~oooao--
DDOD THIS 1s A C ~ I V I ~ Y .D. ************
EEEE T H I S I S A C T I V I T Y .E. *ttt****l
FFFF THIS I S A C T I V I T Y .Fm ttttt*ttllt
GGGG T H I S I S A C T I V I T Y .Gm tttt*t*t*I
HWHH T H I S I S A C T I V I T Y .H. 00000000000- - - -
1111 THIS I5 A C T I V I T Y -1. 00D00D0---
** CPM L A T E STCIRT A C T I V I T Y BARCHART t ** PAGE: 1- 1

SYMBOLS USED ARE: <*5 C R I T I C A L PATH i CWr UOPK DhY i .:-> TOTAL FLOAT i <.' HOLIDAY- WEEKEND i \^;.UPDATE DATE

AUG 81 SEP e l
1 3 1
34567890123~561L190123456789011234567890123
MTWTFSSHTUTFS5MTUTFSSMTWTFSSMrUrFSSMTUTFSS
---------------------------------------------------------.-
LABL DESCRIPTION CALNDR DAYS -> 10 20 30 40 50 60 70 80 90
---------------------------------------------------------------------------------------------.-.-------------------.....~----~~~~--
A111A T H I S I S ACTIVITY .A m a***********
PPBB T H I S I S ACTIVITY .Be *****I****
CCCC T H I S I S hCTIVITY .C. *Illi**tll*~XIfXl*
DDDD T H I S I S ACTIVITY .D. f*ttt*tl*ttt
EEEE T H I S I S ACTIVITY .Em V%****I**
FFFF T H I S I S ACTIVlTY .Fm ****lX*Ifl*
~ G G OT~IS I
S ACTIVITY .G. 1*1*111**11
HHHH T~IS IS ACTIVITY -H- t*r***t**t*
1111 THIS IS ACTIVITY .Im *8*****

Figure 4-14 Bar-chart computer outputs for the basic scheduling computations on the network shown in Figure 4-12-no splitting allowed.
-
S
108 1 I Basic Topics

the information given is identical to Figure 4-13, but in graphical rather than nu-
merical form.

DISCUSSION OF COMPUTATIONS WITH SPLITTING ALLOWED

The Case 2-Splitting Allowed computational procedure is more complex than

Case 1, and for this reason the algorithm is deferred to Appendix 4-2. This pro-
cedure assumes that activities can be split whenever the combination of con-
straints associated with the activity in question result in early finish and early
start (or late finish and late start) times whose ditference exceeds the activity
&ration. When this occurs, the activity is assumed to split so that it preserves
the continuity of work flow. For example, in Figure 4-1 1e above, the electrical
task must split as shown: 4 days of electrical, 4 days of interruption, and then
the final 1 day of electrical. Note how this split preserves the continuity of
work flow. Just as framing completes its duration of 10 days, electrical recom-
mences for its final 1 day of its overall duration of 5 days. A different split &
assumedjnthe backward pass-to preserve flowjntherev&se dife&n. That is,
?he first portion of electrical would have a duration of 1 day and the last portion
of 4 days. This preserves flow moving backwards through finishing, and then the
1 day initial portion of electrical. An important reselt of this procedure is that
the earlyllate startjfinish times will be--idneti
for an equivalent arrow diagram. (The reader may find it helpful to redraw the
arrow diagram shown in Figure 4-1 l e and perform the basic arrow diagram for-
ward and backward pass calculations to verify this assertion.) It is important to
note, however, that these splitting rules are arbitrary. The actual activity split
can be chosen later by the "foreman" to be at any acceptable point within the
range of the start and finish times computed for the job in question. It should
also be noted that other splitting rules, e.g., split all activities in the middle, may
give different project durations, longer or shorter, from that which results from
the rule adopted here. Without referring further to the algorithm, the results of
its application will be given here to illustrate how it would be used when a com-
puter is utilized to perfom the computations.
Utilizing the same network shown in Figure 4-12 where no splitting was al-
lowed, the results of allowing splitting are given in Figure 4-15. To show that
these results are identical to those obtained for the equivalent arrow diagram,
Figure 4-16 is given with earlyllate startlfinish times and (total) path float times
noted. To compare these two figures, recall that the precedence diagram treats
each task (activity) in totality. For example, activity F is broken into 4 sub-
activities in Figure 4-16. The early start time of the first subactivity is 15, and
the early finish time of the last one is 3 1. These are the only two times shown
for activity F in Figure 4-15. Similarly, the latest startlfinish times are 20 and
38, respectively. A special situation occurs here in that activity F is on the criti-
 Basic Scheduling Comp~tationr 1/09

Two critical paths

I
15 24
'ZZ3,5FSS3& FF5

0 12 3 17

LS LF Project network

Figure4-15 Example network with forward and backward pass times shown--splitti
allowed.

cal path, yet its start time has 5 units (20-15) of float, and its finish time
h t s (38-31) of' float. The critical nature of activity F lies in the fact tha
&d-subactivities (8-10 and 10-15 in Figure 4-16) are critical. This fact is noted
at the top of Figure 4-15, where the critical path is shown to contain F(UR
i.e., activity F has a midportion that is reverse critical. The latter occurs
its predecessor constrains its finish time (FF7) and its successor
start time (SS8).
Comparing Figures 4-12 and 4-15, we note that the p-ct
IessT42-38) when splitting is allowed. This results from the
is required to split, because its late start time plus its
than its early finish time. That is. '

Thus, there must be a 4 unit idle time between the two segments of activity
Figure 4-16 Arrow diagram equivalent to the precedence diagram shown in Figure 4-15--rplitting allowed.
 Basic Scheduling Computations 1111

indicates that none are remired t o split because, in each case, their late start
times plus their durations equals or exceeds the early f i n i s h w e (LS + D 2 EF).
Another difference in Figures 4-12 and 15 is found in their critical paths. In
Figure 4-12, it is a single path, with a duration of 42 units. In Figure 4-15, there
are two critical paths, each having a duration of 38 units. These paths are shown
at the top of Figure 4-15.
To illustrate the use of computers for these splitting allowed computations,
the completed basic scheduling computations for this network are shown in Fig-
ure 4-1 7. The upper portion of this figure essentially gives the results shown in
Figure 4-15 in elapsed working days, and in the lower portion in calendar datks.
The results given in Figures 4-15 and 4-17 are identical. From the above discbs-
sion, it is clear that the equivalent arrow diagram in Figure 4-16, contains more
information about the subactivity schedules than is shown in either Figures 4-15
or 4-17. To recapture this information, the precedence diagram computati&s
are given in bar-chart form in Figure 4-18. Note how the two portions of this
figure give identical activity breakdowns, and earlypate startlfinish and float
times. For example, the peculiar critical nature of activity F was discussed
above. In Figure 4-16, note that the middle 4 units of activity F are critical
(zero float) with startlfmish times of 24 and 28. This is shown in the upper por-
tion of Figure 4-18 by the critical symbol (*)placed on days 25,26,27 and 28.
Also, the last 3 units of F are noted with the noncritical symbol (0), followed by
7 units of slack noted by the slack symbol (-), exactly as shown on the arrow
diagram in Figure 4-16. To recapitulate for activity F with splitting allowed, the
bar chart indicates that itsfirst four units can be started as early as the 16th ddy,
and they have 5 days of float. The middle 4 days are critical and must be per-
formed on days 25 through 28. Finally, the last 3 units have an early start on
the 29th day, with 7 days of float. This figure does indeed contain a large
amount of information with the advantage of having all subactivities shown by a
single bar (line). A longer time will be required to instruct users how to inthr-
pret this chart, however, the benefit/cost ratio appears t o favor its more wide-
spread use in the future.

SUMMARY OF BASIC SCHEDULING COMPUTATIONS

I
The basic scheduling computations have been defined for arrow and node dfa-
grams as the computation of the earliest and latest start and fmish times of eakh
activity; the computation of path float then follows immediately. In this c h a p
ter a simple procedure has been presented to make these computations directly
on the network arrow diagram using special symbols.
As an alternative, one may make them in a tabular manner, separate from d e
network. A tabular procedure is given in Appendix 4-1. In comparison with
computations on the network, tabular procedures are somewhat tedious and are
 OUNER: UDDER T E X T - THIRD E D I T I O N CONTRACTOR: I AM AN E X C E L L E N T CONTRACTOR

COMPUTER S O L U T I O N SHOUING THE ' S P L I T T I N G .

TEST CALCULATED ON P M K A G E BY K C CRANDALL

OPTION CALCULATION
F O R THE T H I R D E D I T I O N OF THE NODER T E X T ON 'CPH I P E R T m .
UPDATE
CALENDhR
MODIFICATION
3AUG81 1 6 : 5 5 : 1 1
3AUG8l 17:04:38
N E T MOD
RUN NO. 1
1
E.
A C T I V I T Y SCHEDULE L I S T I N G -
CALCULATION 3AUG81 1 7 : 4 2 : 3 8

PAGE 1
-H'"
-I

ACTIVITY START FINISH FLOAT

LhBL DESCRIPTION DURATION EARLY LATE EARLY LATE TOTAL F R E E

8 AAhA T H I S I S ACTIVITY 'A'

I BBBB T H I S I S A C T I V I T Y 'Am
CCCC THIS I S ACTIVITY 'C'
ODDD THIS I S ACTIVITY 'D'
8 EEEE THIS I S ACTIVITY 'E'
FFFF THIS I S hCTIVITY 'F.
* GGGG
S HHHH
THIS
THIS
I
I
S
S
ACTIVITY
ACTIVITY
'G'
'H'
8 1111 THIS I S hCTIVITY '1'

A C T I V I T Y SCHEDULE L I S T I N G PAGE 1

ACTIVITY STClRT FINISH FLOAT

LhBL DESCRIPTION DURATION EARLY LATE EARLY LATE TOTAL FREE

I A A M T H I S I S A C T I V I T Y .On nuc e l AUG 81 1 4 hUG 81 1 4 AUO 81 0 0

I BBBB T H I S I S A C T I V I T Y '8' AUG 81 AUG el 1 9 AUG 81 I 9 AUG 81 0 0
CCCC THIS I S A C T I V I T Y 'C' AUG 81 AUE 81 26 AUG 81 2 8 AUG 81 2 2
nnnn
-..- rurs
.. .
S
I- AC-TI - VITY
- 'om AUG 81 AUG 81 17 AUG 81 17 AUG 81 0 0
t EEEE THIS I S A C T I V I T Y 'Em flUG 81 AUG 01 26 AUG 81 26 AUG 81 0 0
I FFFF THIS I S A C T I V I T Y 'F' AUG e l AUG el 2 SEP 81 9 SEP 81 0 0
* GGGG THIS I S h C T I V I T Y 'G' flu0
AUG
e l
81
AUG
AIJG
el
81
9 SEP
6 SEP
01
01
9 SEP
6 SEP
81
81
0
0
0
0
WHH THIS I S f l C T I V I T Y .H'
I1111 THIS I S A C T I V I T Y 'I' AUG 81 SEP 81 9 SEP 81 ? SEP 81 0 0

Figura 4-17 Time and date computer o u t p u t s for the b a s i c s c h e d u l i n g c o m p u t a t i o n s on the network shown in Figure 4-15-splitting allowed.
 OWNER: MDDER TEXT - THIRD E D I T I O N CONTRACTOR: I M AN EXCELLENT CONTRACTOR

COMPUTER SOLUTION SHOUING THE . S P L I T T I N G m O P T I O N CALCULATION UPDATE NET flOD 1

FOR THE T H I R D E D I T I O N OF THE HODER TEXT ON .CFH % PERT.. CALENDAR 3AUG81 16:55:11 RUN NO. 1
TEST CALCULATED ON PACKAGE BY K C CRbNDALL HODlFICATION 3 A U G 8 1 17:04:38
CALCULATION 3 b U O E 1 1?:4?:38

I 8 1 CPH EARLY S T f R T A C T I V I T Y OARCHART 8 8 8 PhGE: 1- I

SYMBOLS USED M E : CR
i*:,I T I C A L PATH i s.01 UORK DAY i ':-' TOTAL FLOAT i C.> HOLIDAY- UEEKEND i <-> UPDATE DATE

AUG 6 1 SEP 8 1
1 2 3
345678701234567C9012345678?01123456789
bTUTFSSMTUTFSSWrUlFSSMTUTFSSUTUTFS5HTY

L A Q L DESCRIPTION CALNDR DAYS -> 10 20 30 40 50 60 70 80 PO

AAAA T H I S I S ACTIVITY 'A'

BQQQ T H I S I S ACTIVITY 'Q'
CCCC T H I S I S ACTIVITY 'Cm
DDDD T H I S I S ACTIVITY 'D'
EEEE T H I S I S ACTIVITY 'E'
FFFF T H I S I S fiCTIVITY 'Fm
EGGS T H I S I S ACTIVITY .G'
HHHH T H I S I S ACTIVITY .H' 888888111*1
1111 THIS I S ACTIVITY '1' 0000- - - I**
t I 1 CPH L A T E SThRT A C T I V I T Y Q'IRCHART 8 8 I PAGE: 1- 1

SYHQOLS USED ARE: 'R I T I C A L PATH i

<IC <O> UORK DAY i <-> TOTAL FLOAT i C.; HDLIDAY-UEEKEND i C-> UPDATE DATE
Ig.
AUG 8 1 SEP 8 1
1 3
345678?01234567870123456789011~L(45678?
~TUTFSSHTUTFSSflTUrFSSMTUTFSSMTUTFSSMTU S
-.
E
LABL DESCRIPTION-> 10 20 30 40 soCALNDR MYS
do 70 80 so 'a
m
-----.---------------------.--.-.....--.----------------------------
AAAA
QBBQ
THIS
THIS
I S
IS
ACTIVITY
ACTIVITY
'A'
'9'
Itt8tt888888
88----8188888*
s
3
CCCC THIS IS fiCTIVITY 'C' 888ttltl888**t8ttt a
DDDD THIS IS ACTIVITY 'Dm 111*8888~88t
6
EEEE
FFFF
0000
THIS
TH
S
I
THIS
I S
I
S
IS
ACTIVITY
ACT
VIT
IY
ACTIVITY
'Em
-F.
-0'
I81888888
***n*r*-------ttr
~ 1 ~ 8 1 ~ 8 8 1 1
a
HHHH THIS IS ncTrvrrr 'H' T**;X**X***
1111 THIS IS ACTIVITY 'I' 8*88888
114 l 1 Basic Topics

less efficient than computations made directly on the network. A tabular pro-
cedure is included in this text primarily because of its value in helping to under-
stand the logic of computer procedures used t o carry out the basic scheduling
computations. Anyone planning to program a computer for this purpose should
study these tabular procedures.
The computational problems resulting from the introduction of multiple ini-
tial and terminal network events, or the introduction of scheduled times on key
milestone events have been treated in this chapter. Their effects on the compu-
tational procedures are trivial; however, occasions may arise where they can be
profitably applied. The problem of network time status updating was also con-
sidered. Again this results in a trivial modification of the basic scheduling com-
putations, but it is an important procedure in the control phase of project man-
agement. Finally, the computational procedures associated with precedence
diagrams is given, for the cases where splitting of activities is or is not allowed.
In this procedure, activities are treated in their totality; early and late start t i e s
are given for the beginning segment, and early and late finish times for the end-
ing segment of each activity.
After having developed anow and precedence diagram procedures in this
chapter, it is clear that each has some significant advantages over the other. The
uniqueness and simplicity of the arrow or node diagram are very desirable prop-
erties. However, the compactness of the precedence diagram, 9 activities in Fig-
ure 4-15 vs. 25 activities in Figure 4-16, is also a significant advantage for pre-
cedence diagrams. The latter advantage will be greatest in construction type
projects where concurrency of activities is the rule rather than the exception,
and least in some developmental type projects where concurrency tends t o be
the exception. On this basis, one might conjecture that the use of precedence
diagramming will undoubtedly grow in the 19809, but not to the point of com-
plete replacement of the arrow diagram.

REFERENCES

1. Battersby, A., Network Analysis for Planning and Scheduling, St. Martins
Press, Inc., New York, 1964, Chapter 3.
2. Crandall, K. C., "Project Planning With Precedence Lead/Lag Factors," Proj-
ect Management Quarterly, Vol. 4 , No. 3, Sept. 1973, pp. 18-27.
3. Fondahl, I. W., A Noncomputer Approach t o the Critical Path Method for
the Construction Industry, Dept. of Civil Eng., Stanford University, Stanford,
Calif., 1st Ed., 1961, 2nd Ed., 1962.
4. Fulkerson, D. R., "Expected Critical Path Lengths in PERT Networks," Op-
erations Research, Vol. 10, No. 6, Nov.-Dec. 1962, pp. 808-81 7.
5. IBM, Project Management System, Application Description Manual (H20-
0210), IBM, 1968.
6. Moder, J . J., "How to Do CPM Scheduling Without a Computer," Engineer-
ing News-Record, March 14, 1963, pp. 30-34.
 Basic SchedulingComputations 115

7. Montalbano, M., "High-Speed Calculation of Critical Paths of Large

works," IBM Systems Research and Development Center Technical Re
Palo Alto, California, undated report, about 1963.
8. Weist, Jerry D., 'trecedence Diagramming Methodr Some Unusual Cdarac-
teristics and Their Implications for Project Managers," Journal o f Operations
Management, Vol. 1, No. 3, February 198 1, pp. 121-130. I

EXERCISES 1
I
1. a. In Figure 4-6, suppose an activity 7-5 must be added to the netLork
which requires 1 time unit to carry out. Will this change any of the
times computed in the basic scheduling computations? I
b. What time value for activity 7-5 would cause it t o just become critical?
c. Suppose the project represented by Figure 4-6 is the maintenance,of a
chemical pipeline in which activity 0-6 represents the deactivation of the
line. To minimize the time the line is out of service, when would you
schedule this activity? 1

~
2. Redraw the network shown in Figure 4-7 and perform the network cohpu-
tations using the following scheduled times.
a. The main project is scheduled to start at time zero.
b. The activities leading t o event 210 are scheduled t o be completed 12
days after the start of the main project.
c. The scheduled time for the completion of activity 108-211 is 20days
after the start of the main project. I
d. The scheduled time for the completion of the main project (event 109) is
also 20 days. I
3. a. In exercise 2 what is the critical path?
b. In exercise 2 what is the effect of assigning a scheduled time of 16 t o
event 106, or a time of 12 t o event 106?
4. It is possible t o modify the network in Figure 4-7 so that the correct basic
scheduling computations can be carried out by the simple procedurd de-
scribed at the beginning of this chapter. The required modifications are in
the form of dummy type activities with suitable time estimates. For exam-
ple, the addition of an activity 101-210 with a time estimate of 8 would
eliminate the multiple initial events. If the main project had a scheduled
completion time of 17, i.e., Llo9 = 17, what activity and time estimate
would eliminate the multiple end events? I
5. A reactor and storage tank are interconnected by a 3" insulated procesd line
that needs periodic replacement. There are valves along the lines and at the
terminals and these need replacing as well. No pipe and valves are in stock.
Accurate, as built, drawings exist and are available. The line is overhead and
requires scaffolding. Pipe sections can be shop fabricated at the plant. Ade-
quate craft labor is available.
You are the maintenance and construction superintendent responsibl forL
this project. The works engineer has requested your plan and schedule for a
116 1 1 Basic Topics

review with the operating supervision. The plant methods and standards
section has furnished the following data. The precedents for each activity
have been determined from a familiarity with similar projects.
Symbol Activity Description Time (Hrs.) Precedents
A Develop required material list 8 -
B Procure pipe 200 A
C Erect scaffold 12 -
D Remove scaffold 4 I,M
E Deactivate line 8 -
F Prefabricate sections 40 B
G Place new pipes 32 F. L
I Fit up pipe and Valves 8 G. K
J Procure valves 225 A
K Place valves 8 J. L
L Remove old pipe and valves 35 C. E
M Insulate 24 G, K
N Pressure test 6 I
0 Clean-up and start-up 4 D, PI
a. Sketch the arrow diagram of this project plan. Hint: at least three
dummy arrows are required.
b. Make the forward pass calculations on this network, and indicate the
critical path and its length.
c. For obvious reasons, activity E "Deactivate line" should be initiated as
late as possible. What is the latest allowable time for the initiation of
this activity?
d. List the various network paths in decreasing order of criticality.
6. The network plan in exercise 5 is subject t o criticism because failure t o pass
the pressure test could result in several problems. How would you network
this project t o avoid this criticism, and what is its effect on the expected
project duration?
7. Update the network given in Figure 4-6 based on the following activity
progress report submitted at the end of the fifth working day.
Activity Start Time Finish T i m Modifications
0-1 1 3 -
1-2 5 - -
0-3 0 2 -
3-4 3 - -
3-7 2 - -
0-6 5 - -
5-8 - - Activity duration
estimate increased to
four days for 5-8.

a. What is the current status of this project with respect t o a scheduled

completion time of 15 days?
b. What activities must be expedited t o alleviate the situation found in (a)?
 Basic Scheduling Computations 117 1
8. Referring to Figure 4-7 in the text, trace out the critical paths from end 1
events 109 and 2 11, using only the forward pass computations shown on I
the network.
a. From this information alone, can you say which path is the critical path?
b. Explain why you need the backward pass computations, or a portion of
it, to specify that 210-106-108-211 is the critical path.
~
II
9. Redraw the network presented in Figure 4-5 using the node scheme.
a. Complete only the forward pass computations, and show that they are
adequate t o identify the critical path, using the same procedure de-
scribed in the text for Figure 4-5.
b. Complete the backward pass computations and the path float computa- 1
tions. Check your results against Figure 4-6.
10. Redraw the network presented in Figure 4-7 using the node scheme, and
show that the forward and backward pass, and path float calculations lead
to the same results as obtained with an arrow diagram.
11. Consider the precedence diagram network shown in Figure 4-1 9, with activ-
ity durations noted inside of each node.
/

l
a. Compute the earlyllate startlfinish times for each activity, assuming that I
no splitting is allowed, the ES time for the project is zero, and the LF for
the project completion is 27 (let L F = EF = 27, called the zero slack con-
vention). Note that INITIAL TIME = 0 must be utilized to obtain the
I
correct ES time for activity C, and the TERMINAL TIME = 27 is re-
quired to obtain the correct L F time for activities F and G. I
b. Re eat part (a) for the case where splitting is allowed on all activities.

i
c. Id ntify the critical path in parts (a) and (b), note the differences in the
co puted times, and indicate the type of criticality for each activity.
d. Is $putting of much significance in this network?
12. In thd section called "Arrow versus Precedence Method" in Chapter 3, a
projeclt is diagrammed in two forms. The arrow diagram is shown in Figure
I
118 1 I Basic Topics

3-1 2 in the form of an early start time bar chart. The equivalent precedence
diagram is shown in Figure 3-1 3.
a. Compute the earlyjlate startjfiiish times for each activity in Figure 3-13,
assuming splitting is allowed. Compare your results with the arrow dia-
gram in Figure 3-1 2.
b. Repeat part (a) assuming no splitting is allowed, and compare the results
with part (a).
c. Draw the precedence diagram for the entire project.
d. Compute the earlyllate starttfinish times for the network in part (c), as-
suming splitting is allowed.
e. Repeat (d) assuming no-splitting is allowed.
f. Compare your results in parts (d) and (e), and compare them with the
bar-chart output from K. C. Crandall's computer program (reference 2),
given below in Figure 4-20.

JUNE JULY
1972 1972

NTWTF MTWTF MTWTF NTWTF MTWTF MTWTF MTWTF MT

22222 233 11111 12222 22223 11
2 3 4 5 6 9 0 1 1 2 5 6 7 8 9 2 3 4 5 6 9 0 1 2 3 67890 3 4 5 6 7 0 1

1 5 10 15 20 25 30 35
1
2
BOT R E I N F I R S T FLR
M E C H ROUGH I S T F L R
**XXX ----- **
3 E L E C ROUGH 1 S T F L R
*tt
*- +*t*t
xx-.- -----
.---- ----- -
4 TOP MESH F I R S T FLR x--- --xx- ----- ----- ---
XX--- ----- ----- ----- --
5
6
POUR F I R S T F L R
B O T R E I N SECOND F L R XXXXX ----- -----
M E C H ROUGH Z N D F L R
E L E C ROUGH 2 N D F L R
T O P MESH SECOND F L R
POUR S E C O N D F L R
BOT R E I N THIRD FLR
M E C H ROUGH T H I R D F L R
E L E C ROUGH T H I R D F L R
TOP MESH T H I R D FLR
POUR T H I R D F L R

* C R I T I C A L PATH
X N O N - C R I T I C A L SCHEDULE
- SLhCK (FLOAT)
Figure 4.20 Computer bar chart of precedence network.

13. Develop an example t o illustrate a precedence diagram activity that is Bi-

Critical (BC); that is, an activity that would cause an increase in project
duration for either an increase or a decrease in its duration.
Hint: Construct an activity that is reverse critical as a result of an active
predecessor constraint of the finish type (FF o r SF) and an active successor
constraint of the start time type (SS or SF). At the same time add con-
straints that make this same activity normal critical.
14. a. Complete the forward and backward pass calculations on the precedence
 Basic Scheduling Cornputstiom 119

eakh activity.

D F
Test & debug SF Documer t
program program
6 12
A
system specs. - B

program

-
8 12
C E
Collect Run
system data prograrl
4 6
APPENDIX 4-1
ALGORITHMIC FORMULATION
OF BASIC
SCHEDULING COMPUTATIONS

The hand-methods of performing the basic arrow diagram network scheduling

computations taken up in this chapter placed numerical entries directly on the
network. This is by far the most efficient hand method of carrying out these
computations. If, however, networks are to be updated frequently, or if in-
volved questions pertaining to resource allocation or time-cost trade-offs are
raised, then the basic scheduling computations must be carried out many times
on modified input data. For example, in the resource allocation procedures de-
scribed in Chapter 7, the basic scheduling computations may be repeated hun-
dreds of times in the process of allocating resources to the project activities. In
such cases, it is clear that the efficient computer processing of netw0rks.i~very
important. In Chapter 11, the problems associated with the use of computers
are considered from a user's point of view. This appendix will deal with the
computer processing from the programmer analyst point of view.
Computer logic dictates that the basic scheduling computations must be car-
ried out sequentially in some sort of tabular or matrix form. Thus, t o introduce
the subject of programming a computer to carry out these computations, tabular
methods of hand computation will be described. This will be followed by the
presentation of a computer flow diagram to carry out the basic scheduling com-
putations for arrow diagrams. For computer methods t o carry out the basic
scheduling computations on node diagrams, the reader is referred to Chapter 7
of the second edition of this text, or to the reference by Montalbano7 given at
122 1 I Basic Todcr

tivities according to increasing i or j numbers. Fulkerson4 has given the follow-

ing simple procedure to accomplish this end:

1. Number the initial project event (event with no predecessor activities) with
1. (If there is more than one initial project event, they should be num-
bered consecutively in any order.)
2. Delete all activities from the initial event(s) and search for events in the
new network that are now initial events; number these 2 , 3 , . . . , k i n any
order.
3. Repeat step 2 until the terminal project event(s) is numbered.

The network originally given in Figure 4-5 is reproduced in Figure 4-21; it has
its events numbered in topological order. The information describing this net-
work is also given in topological ordered matrix form. Each entry in the matrix
corresponds to an activity in Figure 4-21. For example, in the row labeled initial
(predecessor) event 1, we see entries in columns 2,4, and 7. These are the final
(successor) events of the three activities bursting from event 1. The entries of 2,
2, and 1 give the estimated duration times of their respective activities, 1-2, 1-4,

Final Event i Earliest

Event
Initial Time,
Event 1 1 2 3 4 5 8 7 8 9 Ed

1 - 2 2 1 0
2 - - 4 2
3 - - - 1 6
4 - - - - 5 8 2
5 7
6 - - - - - -4 3 11
7 - - - - - - - 3 1
8 - - - - - - - 5 10
9 - - - - - - - - - 15
'

Latest Event
Time, L, 0 7 11 2 8 12 7 10 15 -
Figure 4-21 Arrow diagram and matrix representation of a project netwbrk.
 s
Algorithmic Formulation of Basic Scheduling Computations

and 1-7. ashes have been placed in the lower left hand portion of this
starting w'th the diagonal cells, to denote that activities corresponding
cells are n t possible in a topologically ordered matrix. Blanks in the upper rjght
hand port.on of this matrix indicate that whiie the corresponding activity is bos-
sible, it is Cot present in the particular network represented by the matrix. 1
the computations by using the matrix in Figure 4-21 to carry
of the earliest and latest event times given by equation
El = 0 in the first row of the column giving the Eils.
entries in the column headed by j = 2. There is
activity in the network precedes event 2. We
in the last column of this same row, E l =
in the second row of the Ei column.
we reach column 6, which is the
E, + D 3 , , = 7 and Es t D,,,

E 6 = Max (E, t D , , , = 7 , E , + D 5 , , = 1 1 ) = 1 1 .
i=3,5

values of Li, we start by arbitrarily letting L, =E9 = 15,

convention described
8. There is only
in the Lj row for column 8. Again, we proceed
this mannkr until we reach row 4, which is the first row that
than one efitry. Here we note D4,, = 5 and D4,, = 8, so that L4 becomes
I
L 4 = Min ( L S - D 4 , , = 3 , L 8 - D 4 , 8 = 2 ) = 2
j = s, s

note that L 1 = E, = 0,which is a check on the accuracy of our

we let L9 = E, = 15.

e and latest activity start and finish times and the total
float can ow be computed in a straightforward manner using
through (7 . These results have been given previously in Table 4-2
different e ent numbers) and hence will not be repeated here.
illustrated in Figure 4-21 readily

g the basic scheduling computations for

exceedin& interesting exercise, because it permits ingenuity to be
great degrde. Rather than following directly the computational
trated in $igure 4-21, a more general approach will be taken
minimum &f requirements placed on the input network data.
assumptiod that i < j for all i - j, and the requirement of a
124 l l Basic Topics

START I
t
Count number of activities
ending in each mode

1
Activity EST is the maximum of:
a) node 111 early time
bl Activity's earliest xheduled

+
start time or its actual stan time

Activity EFT =
activity EST+ duration
(done by date subroutine)

I
NN = activity's successor

+
node number JJJ

Early time of node NN is

the maximum of
al current early time of
node NN

+
b) activity's EFT

Subtract 1 from count of

node NN, and save order in

1
which the activity entered
the solution pet

SetJ=J+l

in the network?

in solution set equal to

NA C R
TO BACKWARD PASS

Figure 4-22 F l o w diagram of logic for the basic forward and backward pass calculations.
 Algorithmic Formulation of Basic Scheduling Computations 125

BACKWARD PASS

I
Set all node late times
t o required project
completion date

KLth activity's LST =

a) activity's actual
stan time, or
b) activity's duration
subtracted from the
minimum of the activity's
scheduled finish time or
the late time of the
activity's successor event
(by date subroutine).

(111) late time as the minimum of:

node's current late time

Set I = l + l

Figure 4-22 Continued.

numbers 101each activity will not be made. The network must, of

and hence no duplication of event numbers is allowed.
to indicate the presence of
in the computational procedure to terminate the

such a computational procedure is given in

presence df a loop in the network is detected by noting
been adde/d to the solution set in the forward pass
through the list of network activities. One
pass cornbutations are greatly facilitated
the activijies in making the forward pass
activities in the reverse order in the backward pass. The computer code for1 this
en in FORTRAN-IV,is given in the second edition of this text.
APPENDIX 4-2
ALGORITHM FOR BASIC
SCHEDULING COMPUTATIONS
FOR PRECEDENCE
DIAGRAMMING WITH
SPLllTlNG ALLOWED

This algorithm, which allows activity splitting, is applied in the same manner as
described above for the algorithm where no splitting is allowed. That is, it is ap-
plied to each project activity in topological sequence, with the same definitions
of the terms INITIAL TIME and TERMINAL TIME. There is one major differ-
ence, however, and that is the requirement to "evaluate" each activity for com-
putational purposes, to see if splitting is potentially required. In this evaluation,
an activity need only be considered as having a maximum of two segments, even
though it can be shown that an activity can subdivide into many segments. For
the precedence diagram network in Figure 4-15, it is shown in Figure 4-16 that
activities B, D and F split into 3 , 3 , and 4 segments, respectively. However, only
two segments need be defined in either the forward or backward pass calcula-
tions to follow.
If an activity is "split" during the calculation process, it is necessary to note
the number of days in the first segment as it could impact the early start of fol-
lowers of the split activity. That is, in those cases where there is a start-to-start
constraint and the first segment of the predecessor has fewer days than required
by the start-to-start relation, the follower cannot start until after a portion of
the second segment is complete. In this case, the start of the follower will be
delayed by the amount of the idle time between the split segments of the
predecessor.
The issue is then to know easily when an activity is "split." There are two
cases that can force splitting; both set the early finish of a follower at a date
126
 Algorithm for Basic Scheduling Computations 127

later t h the early start of the follower plus its duratiop. The fust, and bb far
GiiG :ommon. case is when a finish-finish factor determines the earlv finish
of follo :r ( j ) . This situation occurs in Figure 4-1 l e where FF= I
framing ~delectrical causes EF = 11 for electrical, which is greater than
D=2+ = 7. The result is that electrical is split into two segments with ad idle
gap of - 7 = 4 time units. The last segment has a duration of 1 to preberve
the con uity of work flow, since FF = 1, and the first segment has a durbtion
of5- 1 4. The duration of the first segment will be designated by orj, alnd is
comput as follows, for the case where EFj is set by EFi + FFil:

The r ond situation where an activity can be split during the forward tera-
tion inv ves the SFij factor. When this relationship exists between two a tivi- i,
ties, t h e redecessor can cause the follower to have an early finish greater than
ESi + D The EF of the follower in this case is:

When tl condition establishes the early finish of an activity, _thelength df the

-
first seg :nt must once again be determined toj~_ese~e_work o-lf

The i
lowers 1
factors.
!
ual value of aj is important in the evaluation of the early start o fol-
activity (0,when the relation to such followers involves start-to start
4s in all network evaluations, the forward iteration uses local Aaxi-
mums t establish early start and finish values. When the relation
tivities :ludes start-to-start factors, the potential value for the
followe s related to the early start of the predecessor and
of prog is required to have been completed on the
progres! SSij) may be greater than the number of
of the :decessor, and if this is true the potential early start of the follbwer
must a( bunt for the additional time units still remaining to be accomplihed.
This is me by the following equation: Potential ESI = EFi - Di + SSiP
evaluati I effectively creates a pseudostart for the predecessor, as though
were nc ulit segments, and the potential early start of the follower will be
on the >per number of completed time units on the predecessor. This ditua-
tion occ rs in Figure 4-15 at activity F. Here, the E f i = 31 time is set bk the
F h F= factor between activities E and F, i.e.,
31. Th results in an a~ value of 4 for the initial
DF-A F = 11 - 7 = 4. But the SSFG = 8 factor between activities F
calls for he first portion of activity F to have a duration of 8; hence
lem req ring a pseudo early start. If activity F started at its ESE = 15
ran for time units, then it would have to be interrupted for 5 time
128 I 1 Basic Topics

the last portion of F could be started and run for the remaining 7 time units.
This interruption is required in order to satisfy the FFEp= 7 factor. The 5 time
unit delay is accounted for in the pseudo early start calculation for activity F,
when evaluating the early start time of activity G. That is, ESG = EFp - DF +
SspG= 31 - 11 + 8 = 28, which is 5 time units greater than taking ESG = ESF +
SSFG=1 5 + 8 = 2 3 .
The backward iteration evaluates late starts and late finishes and has the same
complication as the forward iteration in that activities can also split during these
calculations. Basically, an activity splits when its late start is set less (earlier)
than the late finish minus its duration, that is, LSi < LFj - Di. The situation is
completely analogous to the setting of early finish in the forward iteration.
Once again the computational procedure must recognize this situation and re-
cord the duration in the second segment (Pi) which is thereby created.
The complete computational algorithm for the case where activity splitting is
allowed is as follows.

Forward Pass Calculations-Splitting Allowed

Apply the following 3 steps to each project activity in topological order.

STEP 1. Compute the early start time (ES,) of the activity (j)in'question. It
is the maximum (latest) of the set of start times which includes the
INITIAL TIME, and one start time computed from each start-time
constraint of the form FSij and SSii, going to activity j, from prede-
cessor activities indexed by i.

PNITIAL. TIME

IEF, + FSii
1
(for each FSii constraint)
I
ES, = MAX
all i
EFi - Di + SSij (for each SSi, constraint with
cui < SSii)
t
STEP
I
ESi + SSii (for each SSii constraint with
cui 2 SSU,or where cii
was not required)
1
2. Compute the early finish time (EFi) of the activity ( I ] in question.
It is the maximum (latest) of the set of finish times which includes
the early start time plus the duration of the activity j in question
(ESi + Dj), and a finish time computed from each finish-time con-
straint of the form FFii or SFii, going to activity j, from predecessor
activities indexed by i.
 Algorithm for Basic Scheduling Computations 129

EFi + FFii (for each FFii constraint)

EFi - Di + SFii (for each SFii constraint with
f f j < SF;,.)
all i
ESi + SFii (for each SFij constraint with
ai 2 SF&, or where ai
was not required)

STEP 3. If EFi > ESi + Dj, then compute a for activity j as follows:

Di - FFij; if EFj was set by an FFii constraint

~
= Di - SF;;; if EFj was set by an SFii constraint

Backward Pass Computations-Splitting Allowed

Apply the following 3 steps to each project activity in reverse topological orddr.
~
STEP I . Compute LFi, the late finish time of the activity (i) in question. It
is the minimum (earliest) of the set of finish times which inclu es
the TERMINAL TIME, and a finish time computed from each fin'sh-
time constraint of the form FSii or F&i, going from activity i, to
successor activities indexed by j.
I
LSi - FSii (for each FSii constraint)
LSj + Dj - FFii (for each FFij constraint with
LFi= MIN
alli P j < FFii)

I
LFj - FFij (for each FFij constraint with
Pj 2 F h , or where
Pi was not required)
STEP 2. Compute LSi, the late start time of the activity (z) in question.
is the minimum (earliest) of the set of start times which includes
latest finish time minus the duration of activity i (LFi - Di), anh a
start time computed from each start-time constraint of the form idsij
or SFii, going from activity i, to successor activities indexed by j. 1
130 l l Basic Topi-

ILSi - SSii (for each SSijconstraint)

I
LSj + Di - SFii (for each SFiiconstraint with
LSI = MIN
allj pi <SF;)

I
LFj - SFli (for each SF constraint
Pi 2 SQ,ii, or where
Pi was not required)
STEP 3. If LSi < LFi - Di, then compute 0 for activity i as follows:

-S i if LSi was set by an SSijconstraint

,
P-
Di - SF!.. if LSi was set by an SFii constraint

Example Problem

The example previously given in Figure 4-15 shows the results of applying the
above algorithm during the forward and backward pass calculations, for the case
where splitting of activities is allowed. Some of the calculations are shown be-
low for the forward pass.

Activity A

ESA = {INITIAL TIME = 0) = 0

EFA=EsA+DA=o+12=12
~ unnecessary since EFA= 12 is not greater than ESA +
Computation of c r is
DAz12)

Activity B

ESB = MAX
INITIAL TIME = 0
ESA+SSAB=0+3=3
} =3

EFB= MAX
A I
(Since EFB > ESB+ DB, calculate crB)
EsB+DB=3+10=13
}=I7
EFA+FFAB=12+5=17
 Algorithm for Basic Scheduling Computations

Activity D

EFD = MAX
B I ESD+DD=O+12=12
}=I5
ESB+SFBD=3+(2+10)=15

(Since EFD > ESD + D D ,calculate cq,) o . * , - , ~

.-

a D = D D- ~ ~ ; , = 1 21-0 = 2
- -c
Activity C

ESc = MAX
INITIAL TIME

ESA+SSAc=0+3=3
=0
} =3

EFc = MAX
D IESc+Dc=3+18=21
EF,-DD+SFDc=15-12+(7+14)=24

Note: The formula EFD - D D + SFDc was required since

}=24

( Y ~ = ~ < S F ; ) ~ = ~ .
(Since EFc > ESc + Dc, calculate ac)
o l , = ~SF:^
~ = 18- 14=4
etc.
A portion of the backward pass calculations are as follows:

Activity I

LFI = { TERMINAL T I M E = 38) = 38

LSI=LFI- D 1 = 3 8 - 7 = 3 1
(Computation of PI is unnecessary since LSI = 31 is not less than LFI - D I
Activity F

LFF = { TERMINAL T I M E = 38) = 38

132 1 I Basic Topics

(Since LSF < LFF - D F ,calculate OF)

flF=DF-SSFG=11-8=3

Activity C

LFc=MIN
F, H
i
TERMINAL TIME = 38

LSF+DF-FFCF=20+11-5'26 '26
LFH - FFcH = 3 5 - 3 ~ 3 2
Note: Middle equation was required since OF = 3 < F k F = 5 .
LSc=MIN {LFc - D c = 2 6 - 1 8 = 8 ) = 8
I
F, H
(Computation of Oc is unnecessary since LSc = 8 is not less than LFc - Dc = 8 )
etc.
These computational results are shown in Figure 4-15. The two critical paths
through this network are shown at the top of the figure.
PROJECT
COST CONTROL

From the preceding chapters it should be clear that PERT/CPM and precedence
diagramming are time-oriented methodologies. They permit the development of
time-based plans and schedules for projects in great detail, and the monitoring of
actual versus planned times. The basic procedures described in these previous
chapters have, in fact, often been termed "time-only" methods because they do
not explicitly consider possible constraints on the available resources (e.g.,
money, men, and equipment) which may be needed for completion of project
activities. While specific resource requirements (e.g., number of men) are im-
plicitly assumed in deriving the time durations associated with activities in the
network diagram, resources available are ignored.
Although the developers of CPM included a provision for activity time/cost
trade-offs, the scheme was used only as a means of determining the "best" (in
terms of minimum total cost) activity times for use in scheduling the project.
That scheme, which is decribed in Chapter 8, was not designed for complete
project cost control and is not in any sense a cost accounting/control system.
In spite of the time-only orientation of the basic PERTICPM scheduling
methodologies, it was recognized very early on that by adding activity costs to
the network diagram a potentially powerful means of improving project cost
planning/control could be obtained. Conceptually, this addition was not diff-
cult, since none of the basic network scheduling rules were changed. For ex-
ample, consider the simple activity-on-node network in Figure 5-1, previously
shown as an arrow diagram in Figure 4-6. In addition to time durations for each
 $1000- Total cost
c-Activity number
4 t D u r a t i o n . weeks

Figure 5-1 Illustrative network with costs of performing each activity.

activity, the estimated total cost of performing the activity is shown beside each
node. The total activity cost is assumed to be spread evenly across activity dura-
tion, so that a per period cost can be calculated for each activity. Performing
the basic scheduling computations on this network will produce the earliest and
latest start and fmish times for each activity as shown in Table 5-1, which also
summarizes the duration and cost data for each activity.
If all activities are assumed to start at their early start (ES)times, and costs
are summed cumulatively across time periods, the ES cumulative cost curve
shown in Figure 5-2 will be obtained. Similarly, by starting activities at their
late-start times the LS curve shown will be obtained. If the activity costs are
the total costs associated with this project the region between these two curves

Table 5.1. Time and Cost Data for Illustrative Network

Duration. ES LS Total Cost, Cost per

Activity Weeks Time
-
Time Dollars Week

0 5 $1400 $700
0 0 1000 500
0 6 0 0
2 7 6400 1600
2 3 2500 500
2 2 8000 1000
1 7 1800 600
6 11 0 0
7 8 800 200
10 10 3500 700
11 12 4200 1400
 Project Cost Control 135

Figure 5-2 Cumulative cost cuwes for network in Figure 5-1. I

i
is the area of possible project budgets. A schedule to achieve a particular bu get
in this region can be determined by starting activities at specific times bet een
their ES and LS times. For example, in some cases it may be desirable to h ve a
relatively straight line of cumulative costs from project start to fmish (i. . a
uniform rate of expenditure). This can be approximately achieved by jughhg
the scheduled start times of activities within their allowable limits. arehere
also other, more systematic, procedures for choosing the activity start times
which meet such a uniform rate of maximum per-period expenditure constr8int.
These procedures are covered in Chapter 7, which discusses limited-resohrce
scheduling procedures. These procedures can be applied t o the cost planhng
problem by treating money as a type of resource whose availability by lime
periods is limited.
As this simple example illustrates, the network and critical path
computations can be used as vehicles for producing useful
project costs. In this case the information produced is
However there are also other important questions having
which can be answered with the use of network-based
discusses some of these procedures, which deal primarily with the folloviring
basic questions:

1. What are the actual project costs to date?

2. How do the actual costs to date compare with planned costs to date?
3. What are the project accomplishments to date?
4. How do the actual costs of the accomplishments t o date compare withithe
planned costs of these same accomplishments? 1
5. By how much may the project be expected to overrun or underrun the
total planned cost?
136 1 I Basic Topics

6. How do the above questions apply to various subdivisions and levels of

interest within the project?
Some examples of the type of project-based systems and procedures which
have been developed to provide answers to these questions will be given in this
chapter. First, however, it is helpful to understand some of the important prob-
lems associated with their development and implementation as well as their
historical background.

BASIC PROBLEMS I N NETWORK-BASED COST CONTROL

While in theory the concepts of cost control based on the project network are
not complex, the design and implementation of a practical cost control sys-
tem is not readily accomplished. The fundamental problems facing the system
designer may be classified as (1) those related to organizational conflicts, and (2)
those related to the necessary efficiency of the system. The basic organizational
problem is the conflict between the project approach of network cost control
and the fundamental approach of cost accounting procedures found most in in-
dustry. This conflict is manifest particularly in the design of input and output
phases of network control systems. The input to a network-based system re-
quires the development of an activity accounting procedure by which actual ex-
penditure data are coded to provide association with activities (or groups of ac-
tivities) in the project network. The output from the system likewise must
be project-oriented to provide project summary reports, organized by time pe-
riod, areas of responsibility, and technical subdivisions of the project.
The efficiency of the system is a problem because network-oriented systems
lend themselves to major increases in the amount of detail available to the mana-
ger. The level of detail is both the promise and the inherent hazard of such sys-
tems, and it is one of the primary tasks of the system designer to achieve the
level of detail that provides the greatest return on the investment in the system.
A network-based cost control system can easily require routine input data in
quantities and frequencies that project personnel find extremely burdensome.
Unless the requirements are reduced and the procedures simplified, the system
will come to an early death. Similar dangers lie in the design of the data process-
ing and output phases of the system. The use of precedence diagramming, which
often reduces the number of network activities by 50 percent or more, offers
some promise here.

Activity Accounting

Basic CPMIPERT systems gained acceptance rapidly because the critical path
concept filled a generally recognized need for improved, formal procedures for
project planning and scheduling. In the cost control area, however, formal ac-
 Project Cost Control 137

1
counting procedures were established long before the beginning of PERT/ PM.
Thus, one of the greatest obstacles t o the use of network-based cost co trol,
has been the difficulty of developing a network-based system compatible
the established accounting system.
Generally speaking, accounting systems in organizations engaged in pr 'ect-
type endeavors are designed to plan budgets and to report expenditures 0th
by organizationalunit and by project. However, the emphasis is on the orga iza-
tional unit accounts (section, department, division, etc.), inasmuch as the objec-
J
tives of the accounting system are summaries of expenditures by the funct~onal
elements of the organization. Where the accounting codes also permit su
ties by project, the project summary represents the lowest level of
reporting available to the operating management. The purpose
path approach, on the other hand, is to provide more detailed
control within the project.
Cost data, like time data, must be applied to project
and recording of expenses both by cost item and by
means that a more elaborate system is required. Not
and figures to deal with, but there are many typical
application of cost data to a project network:

I
1. Electronic testing gear is purchased for use in several activities in the roj-
ect. Should the cost of the gear be assigned entirely to the purch sing
activity, when the expenditure actually occurs, or should it be alloc ted
over the activities involving use of the gear?
2. What cost, if any, should be assigned to the curing of concrete? App oval
of shop drawings? Negotiation of a subcontract?
3. Should overhead be included or only direct costs? If overhead is inclu ed,
is it computed the same way for all activities?
4. How should the costs associated with project management be show , as
activities or as overhead?
1
Such questions arise largely from the fact that basic time-oriented net
and the list of project cost items represent two different sets of data.
the sets largely overlap, many elements of the project which involve costs have
not been shown in the network. This is particularly true of
other overhead expenses. Certain other activities involve no
sume time and perhaps should account for a portion of the
ous answers to these activity accounting problems have
will give some examples of cost control systems
counting later. But first it is important to review
cost control and to understand the concepts of work breakdown structure
and work packages.
138 l 1 Basic Topics

HISTORY OF PERTICOST

As noted earlier, the idea of collecting costs on the basis of network diagram ac-
tivities was recognized very early as a potential means of improving project cost
control. A few manual procedures and computer programs to answer the 4 cost
questions listed earlier were developed by individual CPM and PERT users in
the period 1959-62. In 1962, a major boost to the interest in network-based
cost control was provided by agencies of the U.S. Government. The Department
of Defense and the National Aeronautics and Space Administration jointly is-
sued a manual entitled DOD and NASA Guide, PERTICost Systems Design:
which emphasized the cost control aspects of "PERT-type systems." Several
companies and agencies in the aerospace field had already been working with
various PERT Cost procedures and computer programs, but the DOD and NASA
Guide served to formalize the interest of the government and thus to inititate
active development of the procedures throughout the aerospace industry. By
mid-1963, the use of PERT/Cost procedures had become a requirement in cer-
tain military research and development projects.
One of the key features of PERTICost was the utilization of a "Work Break-
down Structure" (WBS) to show the hierarchy, or levels, of tasks within a proj-
ect, and the definition of "work packages" at the lower or basic levels of work.
The WD-NASA Guide described these concepts as follows:

End Item Subdivisions. The development of the work breakdown structure

begins at the highest level of the program with the identification of project
end items (hardware, services, equipment, or facilities). The major end items
are then divided into their component parts (e.g., systems, subsystems, com-
ponents), and the component parts are further divided and subdivided into
.
more detailed units. . . The subdivision of the work breakdown continues t o
successively lower levels, reducing the dollar value and complexity of the units
at each level, until it reaches the level where the end item subdivisions finally
become manageable units for planning and controlpurposes. The enditem sub-
divisions appearing at this last level in the work breakdown structure are then
divided into major work packages (e.g., engineering, manufacturing, testing).

The DODNASA Guide suggested that the basic work packages might be
formed by the same individual activities, or groups of activities, used in the
PERTICPM network. Figure 5-3 shows, for example, an illustration from the
Guide of the level-by-level WBS and use of network activities as work packages.
The specification of network activities as the basic cost control unit was
widely misinterpreted as meaning that individual network activities should al-
ways comprise work packages. This created numerous problems with respect to
the acceptance of PERT/Cost. There was belated recognition of the fact that it
was often impractical for contractors to budget and report costs by PERTICPM
 Project Cost Control I 139

Structural
161 Elm? mg 70735
161 M n h . mg. 2M637 Summary
161 Mb. 1A321
161 T ~ & W 1T626
C h a r s numberla" activtier
for w a r t package coding

Figure 5-3 Sample work breakdown structure showing accounting identification cc 's.~

network activities. As J. R. Fox, a member of the original PERTICost de lop-

ment team, notes:

. . . the design group intended to make clear that the work breakdown mc-
ture, not the network, was the basis for cost planning and control. In tro-
spect this was far from clear. Confusion arose because many defensl :on-
tractors as well as other government agencies were developing theil )wn
versions of PERTjCost. Several of these versions called for the use of :RT
networks in the budgeting and reporting of costs by network activities re-
quirement that resulted in the collection of vast amounts of very detaile :ost
information.
By 1964 more than ten variations of PERT/Cost existed throughout OD
and NASA, most of which called for costing of PERT network activiti~ and
the submission of detailed cost information on a monthly basis. Contr tors
recognized the impracticality of these systems and created their own P KT/
Cost groups to prepare reports for DOD. These groups operated sepa tely
140 l 1 Basic Topics

from management teams responsible for the actual planning, scheduling bud-
geting and measurement of program performance. Government auditors de-
cided that PERTICost groups were a legitimate overhead expense that could
be charged to the government. Many contractors considered PERT/Cost as
the basis for negotiating higher overhead rates. . . .

Because of the confusion and resistance fo PERT/Cost throughout the defense

industry, the Air Force, in 1964, developed simplified standards by which a con-
tractor's internal cost management system could be measured before the com-
pany could qualify for defense work. These specifications contained some es-
sential elements of the original PERT/Cost system, including the WBS concept,
but did not include detailed reporting of cost information from PERTICPM net-
work activities.

DEVELOPMENT OF ClSCSC

The simplified criteria developed by the Air Force in 1964 were subsequently
developed into an improved set of criteria which were issued under DOD in-
struction 7000.2 "Cost/Schedule Control System Criteria (C/SCSC)." These
criteria retained the advantages of some of the essential elements of PERT/Cost,
and yet also reflected current industry practice. As Fox notes:5

I t became clear from the PERT/Cost pilot tests that cost planning and control
could be based on a system that was used on most large commercial develop-
ment and production programs. This cost information could be based on a
WBS that subdivided the program or project according to the manner in which
work responsibility was assigned. Project work was traced down through sev-
eral levels of work definition to the point where short-term work packages
could be identified as the basis for planning and controlling manpower. Bud-
gets were then established for each short-term work package. Costs were
estimated at every level of the WBS, to arrive at a total cost estimate for the
program or project. As a contractor began a development program, actual
manhours and costs were assigned to the work packages. As work was com-
pleted the contractor could compare estimates of cost for short-term work
packages with the actual man-hours and cost required to accomplish the work.
Thus the contractor would keep a constant check on whether work was cost-
ing more or less than was estimated.

In line with Fox's observation, the DOD C/SCSCImplementation Guides con-

tinued the WBS philosophy of the earlier DOD-NASA Guide, and commented on
work packages as follows:
 Projsct Cost Control
1.'
1
A work package is simply a lower level task or job assignment. " ork
package" is the generic term used to identify discrete tasks which have de-
finable end results. A work package has the following characteristics:

1.
2.
3.
4.
5.
It represents the units of work at levels where work is performed.
It is clearly distinguished from all other work packages.
It is assignable t o a single organizational element.
It has scheduled start and complete dates.
It has a budget of assigned value expressed in terms of dollars, an-
hours or other measurable units.
6. Its duration is limited to a relatively short span of time, or it is ub-
divided by discrete value milestones to facilitate the objective meashe-
I/
ment of work performed.
7. It is integrated with detailed engineering, manufacturing or o her
schedules.
I
In an attempt to increase the flexibility of its cost accounting requiremknts
the C/SCSC Guide included new provisions for establishment of "cost accouhts.3
and clarified the establishment of organizational responsibilities for elemends of
the WBS. As the Guide states:

The lowest level at which functional responsibility for individual WBS

ments exists, actual costs are accumulated and performance measureme
conducted, is referred to as the cost account level. While it is usually located
immediately above the work package level, cost accounts may be ~ocatehat
higher levels when in consonance with the contractor'smethodofmanagemknt.
In addition t o its function as a focal point for collecting costs, the cost/ ac-
count in a performance measurement system is also the lowest level in the
structure at which comparisons of actual direct costs to budgeted costs are
required, although some contractors also collect costs and make compari ons
at the work package level.
While the WBS defines and organizes the work to be performed, the o n -
11
tractor's organizational structure reflects the way the contractor has organized
the people who will accomplish the work. To assign work responsibilitj. to
appropriate organizational elements, the WBS and organizational structure
must be interrelated with each other. This interrelationship may be visualized
as a matrix with the functional organization element listed on one axis and
the applicable WBS elements listed on the other [see Figure 5-41. Fur her
subdivision of the effort into work packages may be accomplished by asskgn-
I
ing work t o operating units.

The WBS concept promulgated in C/SCSC, with its associated

provisions for establishing work packages and cost accounts,
 Project Cost Control 143

tical compromise in the concept of activity accounting for business organjza-

tions, particularly in the aerospace industry. The approach should be applicdble
in many industries where network-based cost control is desired. In the consthc-
tion industry, for example, the lowest level work packages may be identicd or
nearly identical to the bid items in the project. In some cases the work pack ge
would be represented on the network by a single activity; in most cases it wo Idk
t
be comprised of a group of activities within a division of the project, such as he
concrete work for the first floor of a building. The specific work organization
and cost accounts established would naturally vary with the type of projkct,
areas of responsibility of the foremen and superintendents, the capabilitie
the cost accounting system, and other factors related to the particular
and its projects.

MONITORING PROJECT COSTS: VALUE OF WORK PERFORMED

As noted earlier, an effective cost control system not only permits the establksh-
ment of realistic budgets or plans, but also includes provision for rnonitoiing
actual costs. This is done by measuring expenditures against budget and idehti-
fying variances so that corrective action may be taken when required. ~ i h c e
schedule progress is also often monitored by measuring actual activity time her-
formance against estimated times, this means that two separate streams ofi in-
formation, time data and cost data, are flowing to the project managers. The
collection, manipulation, and analysis of even one of these two streams of
data typically present significant problems for management. For example in I
the case of direct labor costs, these are typically recorded on weekly time cards
which require the distribution of each individual's time over the activitied in
which he participated that week. If a network-based cost approach like PEAT/
Cost is used, employee activities must be defined in terms of, or coded t o kor-
respond to, specific network activities. Requirements such as this can add sig-
nificantly to the paperwork burden of project personnel, and can be an impor-
tant factor in the acceptance or rejection of the system.
It is interesting to note the C/SCSC provisions for data collection and rep
ing, since they are directed at easing some of the problems stemming from
added requirements of time and cost progress reporting. The CjSCSC objec
is an integrated timejcost progress monitoring/control system. That is, it dro-
vides information about project time and cost performance from a single data
collection stream. The particular data collected is based on the WBS strucdure
and utilizes the three key cost performance measures.
The key cost measures, defined in (a), can be summarized as:
Actual Cost of Work Perfomzed (ACWP) is the amount reported as actu
expended in completing the particular work accomplished within a &en
time period.
 144 1 I Basic Topics

Budgeted Cost of Work Perfomed (BCWP) is the budgeted amount of cost

for the work completed in a given time period, including support effort and
allocated overhead. (This is sometimes referred to as "earned value of work
accomplished".)
Budgeted Cost of Work Scheduled (BCWS) is the budgeted amount of cost
of the work scheduled t o be accomplished in a given time period (including
support and allocated overhead).

Although C/SCSC requires the collection of these costs by cost account level,
these costs can be collected at any level of effort (i.e., project task, subtask,
work package, etc.). These cost measures are utilized in establishing both cost
and schedule (time performance) variances as follows:
Cost variance = BCWP - ACWP
Schedule/Performance variance = BCWP - BCWS
The use of both cost and schedule variances calculated in this fashion provides
an integrated costltime reporting system which measures cost performance in
relation to the work accomplished, and ensures that both time scheduling and
cost budgeting are constructed upon the same data base.
Figure 5-5 shows a popular graphic report from such an integrated time/cost

Millions
of dollan

percent
overrun +:::-,/-%---/
-
+I5'

0 /
,----I
- I
I
Overrun

Percent
- 1
underrun

Months
behind
+2
+I
-
+3 -
- ,------
__------ < I
Months behind

Months 0 /-- I
ahead -1 - I
1 1 1 1 1 1 1 1 1 1 1 I I I I I
J F M A M J J A S O N D J F M A M
1981 1982

Figure 5-5 Performance report from integrated timelcost control system."

 I

Project Cost Control

I'
145

control system. The Budgeted Cost Schedule curve (BCWS), is essentially

of cumulative costs corresponding to a particular schedule of activity
as illustrated earlier in this chapter. The report shows that budgeted
tures up through October 31, 1981 ("today" on the graph) were $3.6 midion.
Actual costs expended in the performance of project activities (i.e., ACWP~are

costs have consistently exceeded budget for the past several months.

pected. However, if the project is just on schedule,

geted for the tasks involved. The "Value of

displays this information.
The fact that the actual costs curve is

value of work completed, BCWP, ($2.3 million) by $2.3 million, which $ an

overrun of 100 percent. The "Cost Overrun" curve in the lower part
figure is a plot of the calculated percent over- or underrun at any
This percentage is obtained by the calculation:
(ACWP) - (BCWP)
% overrun (underrun)
(BCW
where a negative value indicates an underrun has occurred.
In this example, the Value of Work Completed curve is also below the dd-
geted Cost curve, which indicates that the project is behind schedule. he:/
amount of delay at any particular time is taken as the horizontal distanc be-

!
tween the Value of Work Completed curve at that point in time, and the ud-
geted Cost curve. For example, at the current report date, work budget d at
$2.3 million has been completed. But that same amount of work was orig' ally
scheduled t o have been completed about the middle of August, which eans
that the project is about 2; months behind schedule. The "Months Be 'nd"
Curve at the lower part of the figure is a running tally of the number of m nths
behind. It shows that the project has fallen farther and farther behind sche ule.

he Vlhle of Work Completed is also sametimes referred to as the ''earned value"

!
"earned value of work performed," as noted earlier. For this reason, this approach
termed the "earned value approach" to cost control.
 Project tasks

Engineering

Fabricationlaa'y

Test'inspect

Evaluation

2 4 6 8 10 Months

Figure 5-6 Planned schedule and costs of major tasks, prototype development project.

These cost measurement concepts are also useful for cost control at the task,
subtask, or work package level. For example, Figure 5-6 shows the major tasks
of a planned 10-month prototype development project. Total project costs are
estimated at $1.865 million. Of this, $765,000 is engineering work. Assume
that the project has been under way for 3 months and we wish to know how the
engineering work is progressing.
A separate schedule of the project engineering work is shown in Figure 5-7,
which indicates that 9 major subtasks are involved. The veritcal dashed line in
Figure 5-7 indicates the report date; as can be seen, 3 tasks have been completed,
5 have been started and are in progress, and 1 has not yet been started as of the
reporting date.
Figure 5-8 shows the type of information which might be collected on the
engineering subtasks at the end of the 3rd month, using the cost measures de-
fined earlier (BCWS, BCWP, ACWP). This report shows that each of the 3 com-
pleted tasks has experienced a cost overrun, ranging from slight ($500) in the
case of Material Tests to substantial ($13,000) in the case of System Design.
Given this information, the status of the 5 tasks currently in progress is of
considerable interest. Their status is summarized in Figure 5-9, which interprets
the significance of the cost measure relationships and projects total engineering
costs at completion, based upon progress to date. As can be seen, engineering
work is generally behind schedule and over cost. If the present rate of progress
continues, total cost for these tasks will be $812,000 instead of the originally
estimated $765,000.
The cost measurement concepts illustrated by these examples represent an
important step towards the development of effective, implementable cost con-
trol procedures in project management. While these procedures are today used
primarily on government-related projects, they are equally applicable to private
 R o i e a Cast Control

System design
I

Subcontract specs
I
Material tests I
I
Specs review

Drafting
II 1 - 1
Fablass'y support
I
I
Fablass'y process $130 K
I
I
Test support
I I $loOK
1
Engineering release
and review
I I I I I I I I
1 2 3 4 5 6 7 k
Total planned costs - $765 K
Figure 5-7 Major engineering subtasks, prototype development project.

ENGINEERING SUBTASKS STATUS A T E N D O F M O N T H 3

(thousands o f dollars1

Task Status BCWS BCWP

System design Cornplated

Subcontract specs Started
Material tens Completed
Specs review Started
Drafting Started
Fabrication/asaembly support Completed
Fabrication/assembly process Started
Test support Started
Engineering release and review N o t Started
Total

F i g l r e 5-8 Status report o n engineering subtasks at end o f month 3.

148 l 1 Basic Topics

Task BCWS BCWP ACWP Status

Subcontract specs $132 $82 $84 Behind schedule and over cost
Specs review 22 15 15.5 Behind schedule and over cost
Drafting 97 59 59 Behind schedule but within cost
Fabrication/assembly process 100 63 67 Behind schedule and over cost
Test support 21 21 21 On schedule and within cost
449
Estimate t o complete: -X 765 = $812 K
422
Summary: Engineering work is behind schedule and a cost overrun is occurring.

Figure 5.9 Status summary of engineering subtasks in progress and estimate to complete.

sector projects. The evidence suggests that more and more private-sector com-
panies are voluntarily adopting these procedures.

AN EXAMPLE OPERATING SYTEM: METIER'S ARTEMIS

The ARTEMIS project mangement system developed by the Metier Company

is representative of the wide variety of commercially-available computer soft-
ware "packages" available today. General characteristics of the system are given
in Chapter 11 along with a number of other systems. The ARTEMIS system is
unusual both for its completeness of time scheduling and cost control options
and for its unique software/hardware configuration. It involves a comprehensive
collection of software programs fully implemented on a minicomputer. The
complete system is designed for easy mobility to distant project sites; success
in this regard is suggested by the fact that in 1981 more than 300 separate
ARTEMIS installations were reported operating in 22 different countries. The
software package offers cost management procedures that incorporate the con-
cept of WBS, Work Package costing, and Earned Value measurement specified
in the C/SCSC criteria and illustrated earlier in this chapter.
The demonstration project illustrated here is a waste disposal plant con-
struction project consisting of about 600 activities with a duration of about
24 months. A portion of the project WBS is shown in Figure 5-10; the Organiza-
tional Breakdown Structure (OBS) is shown in Figure 5-1 1. The project has
4 WBS levels and approximately 100 work packages, each containing several
network activities. Each work package has a Work Package Code, an OBS code
and a description. For each work package the following information is stored:
baseline start and finish dates
current planned start and finish dates
baseline budget cost
 WBS Structure
Project Cost Control 1
1 9

~l
1
1
Heater Site
Tanks
house
(1000) (2000) (3000)
I
Other
Piles Ducts
frame structures
(1100) (1400)

1
Reinforce Base &
I
Reinforce
I
Walls
base klcker walls
114541
(1451) (14521 (14531

Figure 5-10 Work breakdown structure for demonstration project.

ACWP this period

ACWP to date
forecast costs to complete
BCWS this period
BCWS to date
BCWP this period
BCWP to date
Work Package percentage complete
I
The baseline data referred to above is calculated as follows: prior to project stkrt
the Original Early Start (OES) and Original Early Finish (OEF) are calculated
for each network activity. The calculated original schedule start and finish dates
are then summarized up to work package level to give the schedule baselihe.
Also, the original budget costs for each work package are developed and dis-
tributed over the original schedule to generate the Project Baseline, which is
the BCWS. These cost and schedule baselines are set as reference points at the
beginning of the project. This information can be displayed in Project Cost
Plan reports organized by either WBS format or OBS format. Figure 5-12(a)
150 1 I Basic Topics

OBS Structure

(OBS level 0)

co
(OBS level 1)
(General site)

Supervisor L
Testing)

Supervisor K
(OBS level 2)
(Garage)

-----
Pile Pile Pile
(OBS level 3)
(21111 I22111

Figure 5-11 Organizational breakdown structure f o r demonstration project.

shows a WBS format report for the Heater House portion of the project; Figure
5-12(b) shows an OBS format report for the tank superintendent areas of
responsibility.
As actual cost and schedule data are collected in periodic update cycles, new
activity schedules are produced and the following calculations and operations
occur:

calculation of new ACWP

transfer of latest dates from network activities to work packages
calculation of percent complete of work packages
calculation of new BCWP

Figure 5-13 is a sample updated activity schedule for a portion of the project,
and shows the type of information produced for each activity, including work
package and organizational section codes. Figure 5-14 is a Slippage Barchart,
 Project Cost Control I
li
51

which shows graphically the original and current schedule for each activity, re-
flecting the updated information on project completions, duration changes, nd 1
changes in network logic.
Figure 5-15 shows Schedule Performance reports produced after an upd te,
organized by WBS and OBS format. In both reports the actual and forebast
I
schedule performance is compared t o the baseline schedule. Predicted schedule
overruns for each portion of the project, based on the latest data, are indicated
on the right side of each report.
Figure 5-16 is an integrated C/SCSC-type Cost Performance Report produced
1
after an update, organized by WBS and summarized t o Level 1. It shows the dur-
rent and cumulative values of BCWS, BCWP, and ACWP for the elements of lhe
WBS separately and in total, as well as the original and latest estimates of cost at
completion. Figure 5-17 shows similar information for the tank superintendedt's
areas of responsibility of the OBS. Figure 5-18 illustrates the type of detailed
cost information available for individual work packages upon request. Schedule
information can also be displayed for each work package.
The cost performance index shown at the bottom of the Cost performaice
Report (Figure 5-16) is calculated as follows:
BCWP
Cost Performance Index = CPI =-
ACWP
Values of the CPI greater than 1.0 indicate costs below budget. The ~ c h e d l l e
Performance Index also shown in Figure 5-16 is calculated as:
1
BCWP
Schedule Performance Index = SPI =-
BCWS
Values of the SPI greater than 1.0 indicate completions ahead of schedule. $he
ARTEMIS system utilizes these indices in combination with its four-color grabh-
ics capabilities to produce three summaly reports for top level managemint.
The first of these, shown as Figure 5-19(a), indicates the movement of the Cost
Performance Index over time and shows at a glance whether the CPI is abovk or
below budget or trending in an unfavorable direction. Similarly, the redbrt
shown as Figure 5-19(b) indicates the movement of the Schedule Performahce
Index over time. The third top level report (Figure 5-20) plots CPI against SPI
over time to show at a glance whether project performance is currently fador-
able, unfavorable or marginal, and whether there are any short- or long-term
trends in any of these categories.
The ARTEMIS example used here is a good example of a complete netw
based cost control system. For example, it utilizies CPM network activitiek as
basic building blocks for the work packages; and it uses periodically-updatedin-
formation from the CPM schedules to produce time-related cost informadlon
I
METIER/ARTEMIS I
DEnONSTRATION O F AN EARNED V h L U E SYSTEM
CONTRACT NUMBER : 1 / 0 6 / 1 9 3 2

L E V E L 1 SUMMARY : HEATER HOUSE

COMPANY : XYZ COnPANY

Ilrn PIES 11'2111 I I I I I I I I I I I I I 1

1200 E/STRWLPE FRME WII I I I I I I I I lBBBBBBBBDB18Bm1 I I I I I
13W OM ETREIMES I R S 2 I I I I I I IWBMBBDBB I I I I I I 1
IlM DUCTS 117772 1 1 BlDBBmBBmBBBBBBBBBBBBBBBBMBBI I I I I I I I I I

(a1 WBS Level 1 Summary.

METIER/ARTEMIS
DEMONSTRATION OF AN EARNED VALUE SYSTEM
CONTRACT NUMBER ; 1/08/1932
COMPANY ; XYZ COMPANY
PROJECT MANAGER : J .CARTER

SUPERINTENDENT : TANKS - D . HARVEY

PILIh4 - l.Q,lM 7W7b I I BMBI)QlnWBll)UlBBM I I I I I I I I I I I

a*IY CIVIL E M S - 1.M 1332sa I I I I WBsBBMlsnMBWalsBlsBWB~8BIsBIIBW I I I I I 1 I
-
lElYOsCm Y I X S E . M l T l H I61WI I 1 Ils8BWB~BB~BBWBBIBII~BB~BSBB~BBIsBBBBBB
1 I I I I I I I

OVERALL WORK SPAN

-- --- -- ----- ---- - 7
IMLII I I ~ B ~ ~ ~ B U ~ ~ M ~ ~ ~ B M B B M B ~I B B I M ~ IB ~ ~I B MI P BI ~ ~ ~ 2.
M B B
(b) OBS format (tank superintendent's areas).
H0
Figure 5-12 Project cost plan reports. 8
B
6

0-
 A C T I V I T Y SCHEDULE PACE I
*.-=
.-. =-======
UASTE DISPOSAL PLANT
===-= =---==== =-=== *.- .....
RUN DATE: I-JUL-79
.==m=.-.

I
PREC SUCC PROJECT UORK DURA EARLIEST EARLIEST LATEST LATEST TOTRL
ACT. ACT. DESCRIPTION SECTION
-
..-
..
. --- ---- ----
PACKAGE TlON START
----- FINISH
- -----
START
---
..
FINISH
---..-
FLOAT
----- FE.
88 COMPLETE R E I N F BASE TANK 5 1
2-
- Z COMPLETE R E I N F RASE TANK P I
20 FORM-~ING WEAn TANK P I
130 S T R I K E UALLS (TANK P 2 )
210 S T R I K E Y I L L S DUCT B
258 FORM DUCT 'e' (TANK P 2 )
2L2 B A C K F I L L DUCT B
132 FORM (TANK P 2 ) ROOF SOFFIT
134 COMMENCE F I X REINF TO ROOF CEN
I56 CONC R I N G BEAM TLHK P 1 1 1 )
262 COMPLETE F I X REINF TO ROOF DUC
158 S T R I K E RE TANK P 1 1 1 ) AND FORM
264 CONC ROOF DUCT
370 DELIY
266 DELAY
$60 COWCRlNGBEAHTANKPL(2)
372 S T R I K E FORM SOFFIT DUCT A
268 S T R I K E FORM ROOF DlJCT .AY
24 F I X FORM TO CTR BASE TANK P i
I62 STRIKE FORM RE TANK P I ( 2 I
-. .--.
.
136 COMPLETE F I X R E I N F TO ROOF FEN
270 B A C K F I L L DUCT A
164 FORB RING BEAH TeNK 5 1 1 1 )
138 CONC ROOF CENT DUCT
158 DELAY
DELAY
S T R I K E FORM CENT DUCT
S T R I K E FORM CTR DUCT ROOF
DELAY
CONC CTR BASE TANK P I
B A C K F I L L CENT DUCT
EXC T R I M h B L I N D TANK P 2
EONC R I N G BEAM TANK S i ( 1 )
COMMENCE EXC T R I M AND B L I N D HH
S T R I K E RB TINK S I ( 1 I AND FORM
S T R I K E FORM CTR BASE TANK P I
COMPLETE EXC TRIM LIND B L I N D HH
COMMENCE R E I N F B I S E S E 4 U DUC
CONC R I N G BEAM TANK 5 1 1 2 )
COMMENCE EXC TRIM b B L I N D HH D
F I X FOR?! CTR BASE TANK S I
F I X SCREED b CONC I N C L BASE TA
COMPLETE R E I N F BASES E b U DUC

Figure 5-13 Activity schedule (portion of project).

 Project Cost Control
I
155

<<<
?'4 1,8 ,
m
2"
k
4,
3 84
88 Eg ZSP
tl /I
ggg Zli
h, 4,
E&S 81
;;; Z!
C I 6 1 .
= I Y1,1
D I Eii
e l 31,

g i g E ,,l
u, 8
- 8 ..1

:~
m

e
X
2:;
=,, ,
+I(

08,
-8,
!j
gg
q
$4;
5EE
aae
_
=

11
.. ,,

sfl gG 53: ii
I I Basic Topica
 SCHEDULE PERFORMANCE REPORT
METIER/ARTEMIS : -------------------
---------------.----========
DEMONSTRATION OF AN ERRNED VALUE SYSTEM
CONTRACT NUMBER ; 1/08/*932

L E V E L i SUMMARY : S I T E O P E R A T I O N S

COMPANY : X Y Z COMPANY

B I 1 I 11 I I I I 1BBmI~)sBB~UIBBBI I I I I
K I I I I1 I I I I IFFFFFFFFFFFFFFFFFFFFFFI I I I 4

-- - - -- -
SUMMARY

30m $111 w111111l111

(b) WBS format.

Figure 5-15 Schedule performance reports.
 1: 3 S T 1:. lii: R P'(3 H Ma N (:: a?. H 1: I:' 1:) l+ -r'
UORK BREAKDOUN STRUCTURE REPORT DATE : 1-IUL-79
.P-
*.
IETIERIARTEMIS : PROJECT NAME ; HASTE DISPOSAL PLANT E'
DEMONSTRATION OF AN EARNED UhLUE SYSTEM CONTRACT TYPE/NURBER 3 F P , i/D8/1938 4
0
T O TAL PROJECT SUMMARY F,
COHPANY I XYZ COMPANY P
CONTRACTOR : BUCKSHEE ENC.
LOCATION : CINCINhTTI

. .~
- .- --- -- -
........................................................................................
CURRENT PERIOD : CUMULATIUE TO DATE AT COMPLETION
.
........................................................................................................

...-. -..- .......--...--.-------..--..-

:BUDGETED COST: : UARIANCE ;BUDGETED COST : ; VARIANCE :
....-.--.....,arT,,A, - - .,
! *,1I,,I1,

:UORK UORX :COST : : UURK WORM $COST : LATEST

ITEM :SCHED PER ,WORK ; SCHEI COST. BCHED PER :UMK : SEHED COST .BUDGETED HEUISE:D VARIbNCE:
: W E D FORM'D :PERF'D : ULE : "LED FORfirD:PERF'D ! ULE ESTIMATE
..................................................................

HEATER HOUSE : 13473 11889 15800 -1584 -3911: 64996 54643 66461 -10353 -LO1B : 1789b4 169227 -111263 :
TANKS : 22996 22166 18324 -850 3822; 103176 102854 ID5446 -622 -2592 : 488i20 409139 -iiOi9 :
S I T E OPERATIONS I PO40 2880 888 840 2592; 7,694 56605 56820 -1089 -215 : 107848 109196 -1448 :

m~~atr~a~~aira*a*rat~a~8;~a~~~~raaiaarai~rrrsttataa~aia1aa~::r:tr~:a~*:rra:*:r~r**a:~***:::~~aasf;xs**aata8arir~rtrt**a:*~****:
TOTAL PROJECT SUMMARY ; 18509 36914 34412 -1595 2502; 286IL6 114102 21D728 -I2064 -4626 : 774932 7976h2 -22730 :
t~rtatf:aaaaaas:a:~aat1~tt:tt11:~t~~~li~~ta~a~tt~atr1a~s~:1:aa::~~1:1i~:ti~:~iaa1r~~~::a*t:a*:*11rxx:~*1~1axm~mma:~:::1:ta:~***8*:
CURRENT TOTAL BUDGET = $774932 ESTIMATED F I N A L TOTAL COST = 1797662
.................... .
.........................
COST PERFORMANCE INDEX =O.98
SCIIED PERFORMANCL INDEX ~ 0 . 9 9

PRODUCED BY I E T I E R I A R T E I I I S

Figures-16 Cost performance report organized by WBS (Level 1 rummaryl.

 C: 0 $3'I' I:' li: I? 1::'LJ 1P M fiN I:: :1; R EE 1 1 CII'I 1 -
FUNCTIONAL STRUCTURE REPORT DATE : 1-JUL-79

METIER/ARTEMIS ; PROJECT NhME : URSTE DISPOSAL PLRNT

DEMONSTRATION OF AN EARNED VALUE SYSTEM CONTRACT TYPE/NUMBER I FP. 1/08/1932
SUPERINTENDENT i TeNKS -
0 . WlRVFI

SUPERVISOR i REINFORCED UORYS - E.RARTIN

COMPANY I X Y Z COMPANY
CONTRACTOR : BUCKSHEE ENC.
LOC*TION : CINCIN*TTI

REPORT PERIOD : OCT-1979

CURRENT PERIOD : CUMULATIVE TO DATE AT COMPLETION

.-.
:BUDGETED COST: : V l R 1 4 N C E :BUDGETED COST ; : YARIeNCE I
R'u
.c,.n.L..-.-...-.' ;--.-.----...:--.....----:fiCTL,AL :.~.-.---.-:
rUORK YORK :COST : : YORK YORK :COST : IATFST
ITEM :SCHED PER :WORK : SCHED COST: SCHED PER :WORK : SCHED COST IBUUC:LTED
RLVISEB YORILINCE'
:ULED FORM'D :PEPF'D : ULE : ULED FORM'DtPERF'D : ULE ESTIMATE

REINFORCE BhSE ( P I )
R I N C BEAM <PI>
REINFORCE YALLS ( P O
REINFORCE BASE ( P 2 >
R I N G BEAM (P2>
REIHFORCE UALLS ( P 2 )
REINFORCE B R M (51)
RING B E M (SIP
REINFORCE UALLS ( S t )
REINFORCE BASE ( 5 2 )
R I N C BEAM (S21
REINFORCE BRSE ( C I >
R I N G BERM (CI)
REINFORCE YRLLS ( e l >
REINFORCE BRSE < C 2 >
R I N C BEAM (C21
REINFORCE UALLS (C2)
:
~ O T ~- LSUPERVISOR : 21039 i2741 14496 -9298 - 1755: 30118 @b712 28776 - 3 6 -2064 : lbOO4O 167755 - 7715 :
s
f
CURRENT BUDGET = S 160040 E S T l M l T E D F I N h L COST = S 167755 0-
Figure 5-17 Cost performance report organized by OBS (for reinforced works supervisor. tanks area).
HETIER/ARTEflIS :
DEMONSTRATION OF AN EARNED VALUE SYSTEM REPORT PERIOD : OCT-1979
CONTRACT NUMBER -
1/08/1932

UORK PACKAGE $ 2121 DESCRIPTION : P I L E S 1P2)

SUMMARY UBS CODE 212 DESCRIPTION : PRIMARY TANK N 0 . 2

UORK PACKAGE SUMMARY

-.=.'=*.-==-=========

DATES :
-----
BUDGET DATES : START = 12-JUL-79 FINISH F lb-NOU-79
CURRENT DATES : START i 17-JUL-79 FINISH i E9-NOU-79

COST INFORMATION
--.
----.---....

CURRENT BUDGET : Sll97b.00 ACTUAL COST TO DATE : SWb91.20 REUISED ESTIMATE : Sl2293.lZ

QERCENT4GE COMPLETION TO DATE 1 76.90

.-=* .==- ---.= -.

COST CODE D E T A I L TO DATE
.-==

ACTUALS TO ACTU4I.S TO
COST BUDGET BUDGET DATE DATE FORECAST FORECAST
CODE MANHOURS * MANHOURS S MANHOURS S

REPORT SHOULD BE RETURNED TO SUPERVISOR : P I L I N G - J.P.LANE

PRODUCED BY M E T I E R / A I T E H I S

Figure 5-18 Sample of detailed cost information for individual work package.
 I
Project Cos3 Control I 61

Cost performance index (CPI)

Against project timescale

CPI = BCWPIACWP
MetierIArtemis graphics
Run date: 10 - Aug - 7 J
1.5 -
1.4 -
= 1.3 -
-3
8
1.2 -
m
.= 1.1 -
z
1.0 .
L

5
0
0.9 -

0.8 -
0
0.7 -

0.6 -
0.5 l Ml lJ l s l Nl lJ l Ml M
l ~ J l s l Nl l J l l l li dl ~

0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0 22.0 211.0
Project timescale XYZ proidkt

Figure 5-191s) Project cost performance index over time.

Scheduled performance index (SPI) MetierIArtemis graphics

Against project timescale Run date: 10 - Aug - 79
SPI = BCWPIBCWS
1.5,

1.4 -
- -
-L
Y)
1.3

2 1.2 -
'0
.-
;E 1.1 -

1.0. Y
2
0.9 -
P
3 0.8 -
-J

0.7 -
0.6 -
0.5 I I I I I I I I I I ~ ~ I I ~ ~ I I I I I
M J S N J M M J S N J M
0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 18.0 20.0 22.0 2h.O
Project timescale XYZ p r o j k t
Figure 5-19(bl Project schedule performance index over time.
162 1 I Basic Topics

Costlschedule cumulative MetierlArtemis graphics

performance index Run date: 10 - Aug - 79
CPI = BCWPIACWP
SPI = BCWPIBCWS
r
Favorable

1.2
1.1 -
-
-+-'- I
0
AU

5 1.0 Ni-2>
\ M
P,F
D/'~
0.9 -
0.8 -
.'.
MT----<~

'' .
O.'
- Unfavorable \
\
Marginal
0.6 - '\\,AP
0.5 I I I 1 I I
0.5 0.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5
Figure 5-20 Combined costlschedule performance index plotted monthly, April-March.

such as the distribution of work package costs over time. It is not necessary to
have such close interlinkages between the CPM scheduling and the cost moni-
toring portions of the overall system, as pointed out earlier in this chapter.
However, as these interlinkages are reduced in number and frequency the po-
tential benefits of improved accuracy and reliability of the information pro-
duced will correspondingly decline.

SUMMARY

The concept of network-based cost control offers to project management a

basically sound and powerful way of relating expenditures to the work done,
a way of reporting what management paid for what it got. Unlike other critical
path techniques, however, network cost control cannot be applied easily to only
selected projects, or only during certain phases of a project. The design and im-
plementation of such a control system involves modifications in established pay-
roll and cost accounting procedures of the organization. To obtain an adequate
return on this investment, the system should be consistently applied to the proj-
ects undertaken by the organization. Furthermore, the full use of network cost
control concepts requires data processing equipment.
In the aerospace industry, these concepts have been developed and imple-
mented on a fairly broad basis, due to the interest of the government and the
general availability of data processing talent and facilities. In the other major
project-oriented industry, construction, there has been less incentive from the
customers and relatively less use of computers. The special field of petrochemi-
 Project Cost Control 1
I

cal facilities construction is an exception to this general rule; developments there

I I

have been at an accelerated pace over the past few years and show no signs o f
abating. With the increasing economies of data processing equipment and skr-
vices, and continuing pressures of inflation and high interest rates, there can be
little doubt that network-based cost control systems will be developed and dti-
lized even more widely in the next few years. I
I I
REFERENCES

1. Control Data Corporation, 3400/3600 Computer Systems, PERT ~ e f e r e n k e

1
Manual, Palo Alto, California, 1964.
2. Archibald, R. D., Experience Using the PERT Cost System as a Managem nt
Tool, presented at the Institute of Aerospace Sciences, Los Angeles, June 21,
dl
1962.
3. Beutel, M. L., "Computer Estimates Costs, Saves Time, Money," Engineer&
News-Record, February 28, 1963.
4. DOD and NASA G u ~ d e PERT
, Cost Systems Design, by the Office of t e
Secretary of Defense and the National Aeronautics and Space Administra-
k
tion, U.S. Government Printing Office, Washington, D.C., June 1962.
5. Fox, J. Ronald, Arming America, Division of Research, Graduate School f
Business Administration, Hanard University, 1974.
6. McDonnell Douglas Automation Company, "Automated Cost Control To s
and The Management Scheduling and Control System (MSCS)," St. Louis,
1974; from onginal work by J. D. Wiest.
L
7. Network Analysis System, Regulation No. 1-1-11, Department of the ~ r m $ ,
Office of the Chief of Engineers, March 15, 1963.
8. U.S. Air Force Systems Command, Cost/Schedule Control Systems riteh ha
Joint Implementation Guide, Oct. 1976.
9. Metier Management Systems Company, Demonstration O f An Earned Val e
System, AR TEMIS, 1980.
i
1
EXERCISES

1. For the network in Figure 5-1, assume that it is decided t o plan that all db-
1 I

tivities will begin at the ES times, except for the following activities, which
are scheduled t o begin at the times listed below: II
Schedukd
Activity Stan
A 3
D 5
H 9
C 4
G 5
l1
Compute the planned cumulative costs for the project and plot the results

I~
on Figure 5-2.

~~
164 l l Basic Topics

2. Using the methods given in Chapter 7, schedule the activities in Figure 5-1
such that the project duration is minimized, subject t o the constraint that
the planned weekly expenditures never exceed $2,500.
3. The ARTEMIS Cost Performance Report shown in Figure 5-16 provides
data on the 3 major elements of the Waste Disposal Plant project (Heater, -
House, etc.). Using the information provided in this report, summarize the
performance status of each of the 3 major elements in the current period,
over the total period to date, and what is indicated as likely at completion.
4. The text lists several problems inherent in the design of network-based cost
control systems having to do with activity accounting procedures. How do
these problems appear to be handled in the ARTEMIS system?
COMMENTS ON
PRACTICAL APPLICATIONS

One might say that critical path methods are applicable to the management of a
project from the cradle to the grave. At each of the stages in the life of a project,
there are a number of areas related to the practical applications of critical path
methods that are not covered in the other chapters but are deserving of atten-
tion. This chapter will treat several of these topics, from the project proposalto
the project updating and control phase.
Special attention is given to techniques in the construction industry, in wdch
many firms employ network methods. Concluding the chapter is an introduc-
tlon to the organizational aspects of project management.
I

PREPARING PROJECT PROPOSALS I

The numerous potential applications of critical path methods in the prepara-

tion of project proposals are fairly obvious and straightforward. These appli-
cations are based on the various project planning and scheduling techniques
discussed in this text, except perhaps that these techniques are applied with
less detail than would normally be used on a project that is definitely to be
executed. If the project is t o be executed, the proposal network serves as a
framework for developing a detailed plan and schedule to be used in carrying
out the project.
In cases where the project completion t i e is specified in the contract, critiial
path methods are useful in determining a project plan which will meet the time
166 1 I Basic Tovies

specification. It may turn out that this plan requires the use of certain time
saving features which add unexpected costs, or risks, to those which would nor-
mally be required to complete the project tasks. For example, if performance
time is extremely critical, "crash" time performance of critical path activities
may be required. In addition, it may be necessary to perform certain activities
concurrently, which would normally be performed in series. The added risks of
the expedited plan over the normal plan must, of course, be considered in pre-
paring the proposal cost estimate.
If the proposal requires consideration of alternate completion times and costs,
then the time-cost trade-off procedures described in Chapter 8 may be appro-
priate in arriving at project costs. Similarly, if the project under consideration
will be competing for a Tied set of resources, then the procedures discussed in
Chapter 7 may be applicable. It may turn out that the proposed project sched-
ule dovetails nicely with the phasing out of current projects, or it may put an
extreme burden on certain critical resources. This type of analysis may greatly
influence the profit margin that management places on the project, or influence
the decision of whether of not to submit a bid on the proposed project.
A number of firms jn various industries are now using critical path methods
routinely, in varying degrees of detail, in preparing project proposals. One large
metal-working firm requires that all internal proposals for capital expenditures
over $25,000 be accompanied by a CPM network showing the project plan and
schedule. Some advertising firms submit to prospective clients networks of
proposed promotional campaigns, showing how the activities involved in product
distribution, space advertising, television and radio commercials, surveys, and
other facets of the campaign are to be coordinated. The implementation of a
computerized management system has proved to be a type of project that is
difficult to schedule and control; now it is not unusual for project networks
to be submitted to management, along with flow charts, proposed report for-
mats, and other elements of proposed projects, and for the networks to be
updated weekly or biweekly to maintain status control over such implementa-
tion projects.

CONTRACTUAL REQUIREMENTS

As mentioned in Chapter 1 and elsewhere, the use of network methods has been
boosted greatly by contractual requirements that network methods be used to
report the plans and progress of projects. This has been particularly true in the
two largest project-oriented industries, aerospace and construction. In the aero-
space industry, where a major customer is the U. S. Government and its various
military and space agencies, the accommodation to network requirements was
widespread and relatively quick. Contractors in the aerospace industry are ac-
 Comments on Practical Applications 1167

customed to extensive Government reporting requirements: the firms are rdia-

tively large and equipped with computers and capable technical staffs; and the
contracts are often on a cost-plus basis. For these reasons, contractors have been
able to meet network requirements promptly. Some projects produced net-
works of several thousand activities in a few months.
1,
In the construction industry, however, the circumstances have been quite dif-
ferent. The Government has played a major role here also, through the Corpsof
Engineers, the Bureau of Yards and Docks, NASA, and other agencies respodsi-
ble for large construction programs. But construction companies tend to be
relatively small, low-overhead organizations without computer equipment or
data processing personnel. Also, the contracts are usually on a fixed-price ba is.
Under these circumstances, which exist throughout the commercial as wellas
1'
Government segments of the industry, requirements for critical path methdds
encounter a variety of difficulties. II
The first problem in this application is to prepare specifications for the nkt-
work reporting system desired. Since the contractors are often not familiar
with network methods, are not administratively geared to handle increases i n
technical reporting, and often must pay for outside assistance, the firms will
take very conseorative approaches t o critical path method specifications. The
authors have seen specifications consisting of only one or two sentences, saying
only that "the contractor shall report progress on the project monthly by
means of a Critical Path Method chart and schedule." Requirements that are
this brief, of course, invite the contractor to submit a five-activity network pr
anything that he thinks might meet the minimum technical requirements o f a
network "chart and schedule." I I
At the other extreme, one constructiou office of a Government agency
pared and repeatedly used standard CPM specifications that covered eight
pages and contained almost 1000 words. This specification attempted to detail
exactly how all the initial updating computations were to be made and how the
computer reports were to be formatted (including the column headings for such
items as an "actual early date" and other unique descriptions). These specifica-
tions were so confusing that only a few specialized computer service bureaus in
the region could satisfy them, and the contractors were dependent upon the use
of these service bureaus. I
For those who wish t o prepare contractual specifications for network rePoit-
ing, the considerations listed below are suggested as guidelines for reasonable and
adequate requirements.

1. Definition of Network method. An available document or textbodk

should be referenced as comprising a definition of the basic technique
to be employed. The terms used throughout the specification should be
consistent with the terms used in the reference.
168 1 I Basic Topics

2 . Level of Detail. The best way to specify the desired level of network
detail is to specify a range of the number of activities to be included. The
range selected, however, should be based on practical experience with the
type of project involved.
3 . Graphic Format. If a particular network format is desired, such as the
activity-on-arrow or the activity-on-node format, this must be specified.
As mentioned above, the best way to do this is t o refer to an available text
and adopt the terminology used in that text to write this specification.
One may also elect to have progress reports submitted in the network for-
mat rather than in computer reports. This may also be specified.
4 . ComputerReport. If computer reports are to be required, the data de-
sired on the reports should be spelled out, item for item. The desired
sequences (sorts) of the reports should also be specified, taking care not
t o require more reports than will actually be used. (See Chapter 11 for
more complete discussion of computer methods.) Normally two output
sorts, one by I-J or activity number and one by total float, should be
adequate.
5 . Updating. As described in Chapter 11 some updating procedures give
misleading or erroneous results. Therefore, it may be necessary to refer-
ence or to fully describe the updating calculation required. The frequency
of updating reports, number of copies desired, whether revised networks
are required, and similar details should be mentioned. In this connection,
it should be noted that a requirement to maintain a network on a time
scale may increase the updating cost.
6 . CostReporting. If the network is also to be used for reporting the cost
of work completed, the specific means of allocating all costs to activities
and of determining the percent completion on activities must be fully
detailed.

IMPLEMENTATION

Generally speaking, the implementation of critical path methods involves the six
steps outlined in Chapter 1 and summarized as follows: (1) planning, (2) time
and resource estimation, (3) scheduling, (4) time-cost trade-offs, (5) resource
allocation, and (6) updating and controlling the project.
However, it should be emphasized that the full application of all the tech-
niques presented in this text is not required in order to accomplish these steps
and obtain worthwhile improvements in the plan and conduct of projects. Nor
is tke size of the project a critical factor in the practical economic implernenta-
tion of critical path methods. Networks sketched on the backs of envelopes
have proved to be useful ways of quickly analyzing and communicating the plan
of a small project, such as the preparation of a technical report. The use of addi-
 Comments on Practical Applications 169

tional techniques, such as the PERT statistical approach and time-cost trade-o s k
may also be applied in concept, if not in full detail, to relatively small projects.
Thus, the implementation of critical path concepts can be considered a matter
of routine management practice, rather than an investment justified only for
large or complex projects. II
The implementation of network planning and control techniques should take
place after a preliminary study has been made to determine how the project
tasks are to be broken down and assigned to key personnel within the organiza-
tion or to subcontractors. In large projects, particularly in the aerospace indus-
tries, this preliminary study is quite important; it fust requires that the overall
mission and performance goals of the system be refined to the satisfaction of the
systems engineer. Then, the functional analysis of the system can begin, which
will lead to the design requirements for the proposed system configuration. The
establishment of the base-line design requirements is a major milestone in the
systems engineering portion of a program definition study. It is at this time that
the formal application of network methods can be made most effectively.
I
Utilizing Personnel Effectively

The preparation of the project network is a job for the key management of tke
~
project, the person or team of persons who know the most about the objectives,
technology, and resources of the project. To conserve the valuable time of these
personnel, the networking effort should be a concerted and concentrated one,
not a secondary activity that becomes drawn out and perhaps never completed.
There is a period of time that management must devote to planning, and the
network should be used as a vehicle for, an aid to, and a documentation of this
valuable effort. Technicians can relieve much of the load on the project man-
agement by transcribing sketches into legible networks, by making computa-
tions, and by making preliminary analyses of the schedule. For this reason,
many organizations have trained young men and women as part-time or full-time
critical path analysts.
Ii
Working with Subcontractors

When subcontractors play a major role in a project, they should play a majbr
~
role in the critical path planning, scheduling, and control. In some cases this
may require group meetings to develop the network and time estimates. How-
ever, in many instances it is not practical to call in a group of people unfamiliar
with network theory and expect them all to contribute effectively to the early
draft of a network. In these circumstances it may be better for the prime con-
tractor to develop the rough draft as far as possible, then call in the subcon-
tractors to comment and add time estimates to their particular areas. It is often
170 1 I Basic Topics

possible for a person to read and effectively criticize a network, even though he
may not have the training or experience to develop a network originally. One
may also elect to hold short courses in network preparation, then ask each sub-
contractor to develop a subnetwork of his portion of the project. The feasibility
of this approach depends not only on the size of the project and the scope of
interest of the subcontractors, but also on how complex are the interrelation-
ships among the various areas of responsibility.
Incidentally, one of the important side effects of critical path applications is
the fact that it brings the subcontractors and the prime contractor together to
meet and discuss the project. The group generally discovers technical problems
and begins to work in advance toward solutions of these as well as in coopera-
tion on the planning and scheduling of the project.

PROJECT CONTROL

Once a project is underway, the critical path network and schedule should serve
as a guide to the accomplishment of every activity in proper sequence and on
schedule. It is in the fundamental nature of projects, however, that activities
will seldom start or finish exactly as scheduled. Therefore, updating the plan
and schedule is an important link in the critical path concept. There are no
rigorous or standard updating procedures in general use, except where com-
puter programs are involved. The procedures offered here are suggestions regard-
ing the general approach, with some specific recommendations for particular
circumstances.
There are four functions performed in the process of updating the schedule
alone (without regard to cost revision). These are:

1. denoting actual progress on the network,

2. revising the network logic or time estimates of uncompleted activities, if
appropriate,
3. recomputing the basic schedule of earliest and latest allowable times, and
4. revising the scheduled activity start times and denoting new critical paths.

Item 2 simply requires the erasure of previous notation and replacement with
the new. The other changes may be categorized as Progress Notation and Re-
vised scheduling.

Progress Notation

A field supervisor or other person close to the actual progress of a project should
be assigned the responsibility of making progress notations directly on a copy of
the network. This is the most reliable way to maintain accurate records for in-
 Comments on Practical Applications

put to the network updating procedure. Requiring the responsible firstjiine

I I
171

supervisor to make these routine daily or weekly notations also helps insure that
the network serves its purpose as a detailed schedule of work.
I
I
Exactly how the notations are made is not particularly important, as long as
they are clearly understood and complete. To be complete it is necessary only
that both the actual start and the actual finish of each activity be recorded.
Percent completion notes are not normally required, except where cost control
data are involved.
On the working copy of the new network used to make these notations, it is
J
often useful to use colored pencils for the notation, using a different cola f o r
each updating period. Then when converting the notes t o computer idput
forms, the person doing the encoding knows to pick up only the green date$, or
whatever was the color of the most recent period.
1
If the node scheme is used, the actual start and finish dates may be marke on
the left and right sides of the nodes, respectively. If the network is computed
manually, the node symbol may be designed to allow spaces for the actual dites
to be recorded.

Revised Scheduling

Hand computation of a revised and updated network requires no special a h -

1
metlc. One merely begins the forward pass computation at the last uncompleted
activity on each path. However, it is important that the first dates computedfor
each path be consistent with the effective date of the computation. That is! no
ES or EF date can be earlier than the effective or "cut-off" date of the progress
report. If the first uncompleted act~vityhas not started (or finished), then its ES
(or EF) date must be recorded as at least equal to the effective date, and the
forward pass is carried on from there.
Hand computation does, however, require some attention to the mechan~cs
I
I1
.
of erasure, lest this humble function become a major problem. Erasure of all
remaining schedule times is required, of course, when a complete recomputation
is to be made. One technique that reduces the erasure problem is to make the
scheduling computations on a reproducible print of the network, instead of on
the original tracing. Erasures may be made on the reproducible print withbut
damage to the arrows and nodes of the network.
Recomputation may be necessary every time a network is updated and/or
II
revised. If the actual progress and expected future progress are very close to
the network schedule, there is no need to make a completely revised computa-
tion. The critical path network protects itself, in a sense, from the need for
many recomputations, even when significant delays or accelerations occur,lfor
the network reveals most of the effects of schedule changes without recordpu-
tations. One need only to refer to the float figures, for example, to see wheiher
I
172 1 / Basic Topics

a specific delay will affect other portions of the project. With this information
plainly visible, it is not always necessary to erase and recompute. Recomputa-
tion of the backward pass is not necessary at all if no changes are made in the
network logic or in the estimated activity duration times. If the scheduling
computations are made by computer, then the revision procedures specified by
the computer program must be used.

Time-Scaled Networks

A "time-scaled" network is one in which the arrows and nodes are located by a
time scale along the horizontal border. With the time scale, the length of each
activity arrow (or its shadow projection on the horizontal scale) represents the
activity's estimated duration. The location of the arrow represents its scheduled
start and finish times. Similarly, the location of each node represents its sched-
uled occurrence time. The arrows and nodes may be located by their earliest
times, latest times, or at selected times between these. Slack is customarily in-
dicated by dashed lines.
The obvious advantage of time scaling is the visual clarity it provides for the
analysis of concurrent activities. Time-scaled networkshave revealed problems of
concurrency (such as the intended use of a test facility by two groups at the
same time) that were overlooked in the analysis of the schedule data in the
tabulated form of a computer output. Time-scaled and condensed networks
are also advantageous in presentations to top levels of management, since the
schedule is communicated more quickly and more emphatically by the graphical
technique.
The disadvantage of time-scaling is not so obvious but is nonetheless impor-
tant. Using manual methods, the maintenance of time-scaled networks through
updating and revision periods is expensive. A change in one time estimate or a
delay in the actual progress on one path can change the location of a number of
succeeding arrows and nodes, necessitating the redrawing of a large portion of
the network. Under contracts that require frequency network revisions, the use
of time-scaled networks could keep a tec cian or draftsman busy almost full-
time, and the networks could be delayed days or weeks in reaching the con-
tracting agency.
Many attempts have been made to simplify network revision, so that t i e -
scaled networks could be quickly and economically updated. Listed below are
some of the materials and techniques that have been tried.

1. Gummed Labels and Tape. Transparent materials with adhesive backing

can be placed and moved about on tracing paper or mylar sheets. Only
high quality materials should be used, lest they blister, peel, or make poor
reproductions.
 Comments on Practical Applications 173

2. Computer-Drawn Networks. As explained in Chapter 11, some softwhre

packages for network processing include features for computer-drawn
networks.
3. Steel Panels with Magnets for Events, and Tape or Chalk Lines for Arrows.
Networks constructed in this way have worked for some applications but
cannot be easily reproduced, even by photography, for distribution.
4. Scheduling Boards with Moveable Bars. Slotted aluminum boards 4 t h
moveable bars in the slots to represent activities have been popular,in
certain applications. The bars are labeled and color-coded.

Time-scaling can be particularly worthwhile where it is done as an aid t o

planning and not intended for use after the project is underway. An interesting
example is shown in Figure 6-1, which is a network for a complex surgical pro-
cedure.' The only purpose of this network application is to help develop an effi-
cient, well-coordinated plan, especially in the critical period between the 30th
and 45th minutes. Once the plan is developed, the network has no further use.
Another example is shown in Figure 6-2. Here three key resources in the
project are summarized by time period, illustrating the utility of time-scaling
networks where there are resource allocation problems. As in the previous ex-
ample, this effort is usually worthwhile only in the planning stage of the project.
In some cases, though, the resource problems are important enough to the time

Figure 6-1 Tirne-scaled network of a kidney transplant procedure.'

i
174 1 I Basic Topics

Figure 6-2 Time-scaled network with resource summary. Activities arescheduled to limit
designer requirements to two and welders to four.

and cost economics of the project that the time-scaled network should also be
used to monitor the project progress. As mentioned above, however, project
monitoring with time-scaled networks is practical, only if there are relatively few
activities and the project management is centralized enough that copies of the
network do not have to be distributed. An example is the overhaul of electric
utility equipment where the network is set up in a "control room" where the
project supervisors gather to discuss changes, revise schedules, etc.
Computer software has advanced to the point that time-scaled charting by
computer is much more practical than manual methods in many applications.
This trend can be expected to continue. At this writing, however, it is still
advisable to be cautious about the expense of maintaining time-scaled networks.

CONSTRUCTION INDUSTRY PRACTICES

Since the construction industry is one of the major users of network methods,
it is well t o consider some of the practical "tricks of the trade" in construction
applications. In an informal survey of firms for the third edition of this text in
1980-81, the authors found that network practitioners in the construction
industry varied widely in their level and style of network use. The larger engi-
neering and construction management firms tended to have planning and sched-
uling departments with trained personnel and computer systems (or frequent
 Comments on Practical Applications 175

access to an outside computer service). The smaller firms often employed sched-
uling consultants on a project basis, sometimes on their own initiative but more
often because the project owner or architect required network-oriented plbns
and progress reports. Smaller scale homebuilders, paving contractors, subcbn-
tractors, and other types of firms may not use the techniques at all or may hgve
had only limited exposure via a general contractor.
The practices of some of the scheduling consultants was of interest beca se 1
of their full-time, professional attention to network methbdology and tdeir
variety of applications. Two examples of such firms are illuminating.
I
DDR International is a consulting firm based in Atlanta, Georgia that special-
izes in construction planning and scheduling for general contractors, owners, dnd
architects.' DDR uses computer processing of networks only when required by
the client, which is seldom. Rather, they employ time-scaled networks to c n-
vey both the sequence and the schedule aspects of project plans. Networks L e
I
prepared on tracing paper with translucent mylar labels for activities. The lablels
have adhesive backing that permits moving them around on the chart at will.
The activities are placed on the time scale according to "scheduled" starts ratKer
than ES or LS times. Updated progress is denoted directly on the network, in-
cluding notation revealing the ahead or behind schedule status of each path. The
activities are not rescheduled, however (meaning a revision in the network), tin-
less a significant change is made in the project plan or schedule.
A major purpose of the DDR technique is to keep the schedule from becoIn -
ing too complicated for client personnel, including managers, field supervisoks,
subcontractors, and architects. Computerized analysis and their tabular reports
are considered objectionable for this reason. The most essential facts of the plhn
and schedule can best be communicated, according to the consultant, by time-
scaled networks based on agreed-upon "scheduled" start times for each activity.
Forward and backward passes are made manually to identify the critical path,
but the computations are not otherwise documented.
For an example of this kind of network presentation, see Figure 6-3. No e It
that the format appears to be activity-on-arrow,sinceeachactivityisrepresentkd
by a time-scaled bar. However, the bars are interconnected by dependency lines
following the activity-on-node or precedence method. The network is actually a
precedence diagram, and leadbag factors are sometimes employed. DDR pdr-
sonnel feel that precedence diagramming of this type is "two or three timks
more effective" than the arrow format.
DDR consultants occasionally assist their clients in resource leveling, whe e
one. two, or three crews of given type may need to be scheduled. This is alwais
1
done manually, usually as the network is prepared originally. They do not offer
network-integrated cost control systems, but they do produce cash requirements
schedules and earned value reports, such as those shown in Chapter 5.
Hanscomb Associates, Inc. is a large international engineering and constru
176 1 I Basic Topia

\\
\
!
8 [\
I
I'
FIRE M
*.
PROTECTION PIlPllC r l l t P l O l l C l l O " 1~1~111C111011

'\
7 '\
A
\
..
_ ~~~_.
..
..
._-
._-
..
..
..
..
.___

! 7

Figure63 Portion of a consultant's network illustrating timescaled activityon-node

format. Courtesy of DDR International. Inc., Atlanta.
 Comment. on Practiral Applications b77

........................................................... 2
TAROET
DES(OI1-I-
1 -- DOMlEWTS S M W h E
.lS,*I~ 11111'1C111~11 l1lC.10" COY*,", ...............
4 JAMSOFI COUNTY JAlL BUlLMNO P'
m u m E D FOR
PRElOLE AND PATRICK 1

Figure6-3 Continued 1
178 1 I Basic Topics

tion management firm. They provide a wide range of semices for the construc-
tion industry, only one of which is project planning and scheduling. Several of
the offices of the firm maintain specialists in network methods. Each office is
independent, however, in the methodology employed. Some favor computer
methods and some favor manual methods. At this writing, the Chicago office of
the firm preferred computer methods, using one of the most advanced and flexi-
ble software packages. The Atlanta office never used computers, favoring time-
scaled network presentations instead. As with some other consultants, the
specialists at the Atlanta office of Hanscomb explained that an overriding con-
sideration in their practice was the communication of clear and essential facts
to their client. To avoid confusion or excessive detail they do not even make
forward and backward passes; the critical path and near critical path is revealed
by the time-scaling process when the network is prepared. For an example of a
Hanscomb (Atlanta Office) network, see Figure 6-4. The network is based upon
"scheduled" start t i e s for each activity, not necessarily ES or LS times. The
flags denote significant start dates or key delivery dates. The solid portion of
the bars denotes completed work.
According t o a senior project manager at Hanscomb, general contractors "do
not want and will not pay for computer outputs." They also do not allocate
resources, employ time-cost trade-off methods, or apply a network-integrated
cost accounting ~ y s t e m . ~
In both of the above examples the consultants stress the necessity for clear
communication with the project management and field personnel, unencum-
bered by details or sophisticated calculations that may confuse the most im-
portant facts of the schedule. They also stress economy and efficiency in
methodology. Other practitioners, however, employ some of the most advanced
computer systems and exercise a greater degree of control over the construction
process. The variation depends upon the types and sizes of projects, the skills
and styles of project managers, the preferences of the consultants (if involved),
and other factors.

HUMAN AND ORGANIZATIONAL ISSUES

While this text deals primarily with network techniques as they apply to plan-
ning, scheduling, and controlling projects, it is important to recognize that
successful project management encompasses a broader body of knowledge. In-
deed, it appears that the development of network methods as a distinctly differ-
ent approach to planning and controlling projects has helped to focus manage-
ment theorists and educators on other differences in projects as compared with
other forms of endeavor. The term "project management" (synonymous with
"program management" in government parlance) has now become a recognized
subdivision of management theory. Libraries often list a number of titles under
180 1 I Basic Topics

the subject of project management, usually in the 658.4 series of Dewey Decimal
codes. Courses are offered in the subject by universities, business firms, and
government agencies.
In addition to the quantitative and graphical methods that aid in project
management, the broader subject now encompasses the human and organiza-
tional aspects that distinguish project-oriented endeavors from the more con-
tinuous or process-oriented endeavors. The successful management of projects
is recognized as requiring different supenrisory skills, different definitions of
authority and responsibility, different criteria for the selection and evaluation
of personnel, different operating policies, and other aspects of management.
The reader who wishes to study these topics in greater depth is referred to other
texts, such as Galbraith? Kerzner? and art in.^
Nevertheless, it is well worthwhile here t o identify and comment briefly upon
some of the key issues in the human and organizational aspects of project man-
agement, in order to introduce the reader to the broader subject and to help
place network methods in perspective.

Type of Business

One of the first factors governing the role of project management in any situa-
tion is the degree of project orientation characterizing the business of the
organization. Here we use "business" to mean the overall line of endeavor,
whether it is run by private enterprise, governmental, educational institutions, or
others. At one end of the spectrum are the businesses that produce a continuous
stream of products, goods, and services. Of course these businesses include
manufacturing, mining, chemical processing, financing, insurance, hospitality,
health care, and many others. These businesses tend to be characterized by high
volumes of repetitive operations, which lend themselves to high degrees of effi-
ciency in the use of labor, materials, and facilities. For want of a better term, let
us categorize these businesses as the "process" type.
At the other end of this spectrum are the project-oriented businesses. Major
examples are the construction industry and research and development. As de-
scribed in Chapter 1, these endeavors are characterized by limited repetition,
uniqueness, and defmed time limits for goals or results.
Between the two ends of this spectrum there are many businesses that consist
of both process and project features simultaneously, or which shift back and
forth from a process mode to a project mode. Examples include job shop manu-
facturing, where product changes occur frequently; advertising and sales pro-
motion, where the goals and the methods undergo frequent revision; and any
business that is frequently expanding, adding facilities, changing products or
services, or developing new concepts while also producing a volume of products
or senrices.
 Comments on Practical Applications 181

The organizations that are formed to operate in the project-oriented business

environment are clearly the ones that must be skilled in project management
concepts. It is their way of life. Not only do they use network methods for
planning and scheduling, but they also employ project cost accounting schemes.
The management structure tends to emphasize the project manager's role, which
is a strong but temporary position of authority and responsibility.
At the same time, most project-oriented organizations of any sue also main-
tain "functional" units which remain more-or-less permanent, such as engineer-
ing, marketing, finance, and distribution. The relationship between the perma-
nent functional units of the organization and the project teams is one that
presents particular management problems. Usually the functional units play
supportive roles while the project teams are primal; yet the project teams are
frequently disbanded or reorganized as projects end and others begin. The func-
tional personnel are often expected to give up or loan key people to project
teams at times, and to accept and employ project personnel who may be tem-
porarily unassigned. If not handled well, these requirements can cause friction
and/or inefficiencies in the use of personnel and facilities. As pointed out by
Martin5 even the organizations that are highly oriented toward the project
mode, such as construction firms or hardware development organizations, must
recognize that each project places a unique set of requirements on the manage-
ment structure, because each project differs in many ways from those that came
before and will come after.
In the business that involves a mixture of process and project modes, there is
likewise a mixed need for the traditional management heirarchy along with
project management schemes. This environment presents the most complex and
unstable needs for organizational structure, and thus the most frequent problems.
In these cases, the general management must be aware of the potential prob-
lems and be skilled in establishing policies and practices that minimize the
built-in conflicts between the goals of the mainstream process functions and the
project goals. These skills include a sound working knowledge of project meth-
odology and organizational structures.

Organizational Structures

There are at least three well-defined alternatives for structuring organizations

that involve a significant degree of project-mode operations. They are:

1. Traditional or pure organization as applied to a project.

2. Line-Staff organization.
3. Matrix organization.
The traditional or pure functional structure is pictured in Figure 6-5 in a
typical form. This is the classical structure for process businesses, but it may
182 1 I Basic Topics

Manager

I
Manufacturing

r I 1
I I
Department
level

Figure 6-5 Traditional or functional organization.

be applied, at least theoretically, to a project within a firm. (Martin calls this

application "completely projectized management.") The simple structure pro-
vides some clear advantages over other organizational schemes, including (but
not limited to) the following:

1. Strong management control over the project.

2. Fast response to changing conditions.
3. Personnel dedication to the project, clear lines of authority and personnel
management.
4. Strong control over project schedules and costs; very suitable for profit
center designation.

At the same time, application of the traditional organizational structure for

process activities to a project-oriented business brings some definite disadvan-
tages. Among them are:

1. High cost of maintaining duplicate functional personnel and facilities for

each project.
2. Difficulty of maintaining technical quality when the functional units are
not united.
3. Difficulty of balancing workloads when projects proceed through life
cycles, phasing in and out, resulting in unreimbursed costs.

For these reasons the traditional structure is not likely to be suitable for
project management unless the project is very large and expected to live for
several years, and the sharing of resources with other projects is unnecessary or
undesirable.
The line-staff structure is much more common, especially where both process
and project activities exist together. As pictured in Figure 6-6 (adapted from
Kerzner5), the project leader or manager is in a staff relationship to the line
managers. In this position the project manager temporarily employs the re-
 Comments on Practical Applications 183

General
manager

Project
manager
I

Manufacturing
manager manager manager

Figure 6-6 Typical line-staff organization. Solid lines represent formal authority; dashed
lines represent informal authority or liaison.

sources of the mainline departments to carry out the project. (See also Youker7
and Stuckenbruck?)
The key to the effectiveness of the line-staff structure lies in the balance of
authority between the project and department managers. In some cases the
project manager is only a coordinator whose prime responsibilities are to at-
tempt to "influence" the department managers to keep up their schedules on the
project and to keep the general manager informed of the status. As explained
by Kerzner and Galbraith, project managers in this role must depend on their
personal competence and their political or interpersonal skills in order to be
effective. (To illustrate this relationship graphically, one might revise Figure 6-6
to place the project manager's box below the level of the functional managers.)
More about this coordination role will be mentioned in the next section.
In some cases, more clearly defined authority may be assigned to the project
manager. Personnel, equipment, and facilities may be temporarily but fully re-
assigned from the departments to the project. Martin calls this the "partially
projectized management" scheme. In these situations the personnel have a
"home" t o return to when the subject is done, and their merit reviews and career
plans continue to be managed by the functional department heads.
Obviously the line-staff approach has some strong advantages:

1. Resources are effectively shared, thus minimizing losses due to idle time or
duplicate effort.
2. Important technical skills continue to be developed and maintained in the
line departments.
3. The project manager can concentrate on the product of the project with-
out being distracted by support functions.

Significant disadvantages also exist. Among them are:

1. A high, sometimes unrealistic degree of cooperation is expected of line and

staff managers, who have different objectives.
184 1 I Basic Topics

2. The general manager's skills are crucial in setting up the proper degrees of
authority and in adjudicating conflicts that arise through the project.
3. Projects conducted under this structure tend to overrun schedules and
cost estimates, due to the lack of clear priorities and full authority for
execution.

Nevertheless, the line-staff structure is by far the most common arrangement

in process businesses and in operations that are mixed between process and
project modes. Overall it is cost-effective and flexible with changing conditions.
Project teams are easily disbanded when the project is finished, and new ones
can be formed fairly readily.
Project teams should be easily disbanded when the project is finished (at least
theoretically), and new ones can be formed as needed. (In practice, however,
project teams often resist dissolution; their tenacity for life can become a prob-
lem. See Benningson.')
Suppose one were to more fully develop the line-staff approach to the point
that there was a strong project manager position for each of a number of proj-
ects, and the project managers were endowed with authority that approximately
equaled that of the functional managers. Essentially, this organizational form
would be what has come to be known as the matrix. Graphically it looks like
Figure 6-7 in its typical form.
In concept, the matrix attempts to achieve the best of both the functional
and the line-staff structures. All the technical and specialist personnel have a
"home" in their functional departments, where they receive project assignments,

General
Manager

Functional
units

+
E
&

-
2
E
.-a
0
L

Figure 6-7 Typical matrix organization.

 Comments on Practical Applications 185

technical training and guidance, and merit reviews. At any one time, a tech-
nician is also assigned to a project and receives specific task direction from the
project management.
Thus the technical staff person has two bosses, which would appear to be a
confusing weakness in the structure. However, practitioners of matrix manage-
ment point out that the multiple boss situation is common; it begins from the
time one is born with two parents, and continues through school days and even
into traditional corporate and government organizations. Even the production
line supervisor in a single-product plant is aware of the separate authorities of
his department manager, the quality control manager, the personnel director,
the union steward, and perhaps others. Thus the two-boss arrangement in a
matrix organization is not unique nor necessarily confusing. (See ~leland.")
The project manager in a matrix enjoys the advantage of more clearly defined
authority and responsibility. He or she should not have to cajole or influence
functional heads in order to carry out the project tasks. The matrix does require
that the project and functional leaders have a mutually cooperative relationship,
but the matrix structure itself can enhance that relationship by its equal recogni-
tion of both sides of the organization, thus providing a unifying set of goals for
the organization as a whole.
Other advantages of the matrix are:

1. Quick response to changing project conditions.

2. Personnel recognize project work as the mainstream activity and thus
develop project loyalty, rather than regarding the project as a distraction.
3. The project manager has full control over the project resources, including
personnel and costs.
4. The functional departments perform essentially as support for the project
teams.

Naturally, there are also disadvantages to the matrix:

1. With its emphasis on project execution, the functional departments tend to

be weaker in certain respects, such as development of the state-of-the-art in
technical fields, or the development of new markets.
2. The structure tends to require more attention by general management and
administrative personnel in order to maintain the proper balance of power,
cost control, and performance control. Vertical and horizontal communi-
cations must be open and freely exercised.
3. Multiple, separate projects are vulnerable to duplication of effort or dis-
parate development of techniques and markets.

Clearly, the matrix structure is most appropriate for operations that are pre-
dominantly project oriented, such as engineering and construction firms, re-
186 1 I Basic Topics

search and development organizations, and the like. However, the matrix has
also found wide application in some process businesses for a variety of purposes.
For a more complete discussion see Cleland.lo
The reader is reminded that this review of the organizational aspects of
project management is offered as a brief introduction to the subject. More com-
prehensive treatment is provided in other texts.

COORDINATING INTERDEPARTMENTAL ACTIVITIES

As indicated above, a project manager in a line-staff or matrix organization will

spend a greater proportion of his time "coordinating" activities than a manager
in a traditional or functional organization. For this reason, this task deserves
some consideration here. L. A. ~ e n n i n g s o nhas
~ developed a network based
analytical tool called TREND (Transformed Relationships Evolved from Net-
work Data) to deal with this problem of coordinating project activities. This
technique involves the concepts of "interdependence," "uncertainty," and
"prestige."
The sequential interdependencies among project activities are conveniently
displayed by the project network. If activity A precedes B, then B's work de-
pends on the work of A. In this case, the integrating mechanisms are plans and
schedules. Now consider the concept of "prestige" and "authority" associated
with the groups conducting activities A and B; then collaboration will be more
difficult." While prestige can be inferred from the position of each group in the
organization chart of the firm, relative authority may be established in terms of
interdependencies, where the party that is dependent is considered to have low
relative authority. So, if activity A precedes B, then B would have low relative
authority. Conflict would be predicted if group B depended heavily on group A,
but group B had much higher prestige in the organization than did group A. In
this situation, the project manager could anticipate the particular need for his
services in a coordinating role.
Finally, the concept of "uncertainty" is important. Lawrence and Lorsch12
state that in typically healthy organizations, individuals and groups that must
cope with high degrees of uncertainty in their jobs (such as basic research or
marketing) are different in important ways from their counterparts who do work
involving low degrees of uncertainty (such as manufacturing or accounting). The
differences in these two groups can be measured in terms of their time horizons,
degree of reliance on formal authority, and degree of value placed on task
achjevements versus interpersonal relations. A surrogate measure of uncertainty
might be the range of activity duration time estimates (pessimistic time minus
optimistic time)* used in the PERT system described in Chapter 9. It is con-
*The coefficient of variation (ratio of the standard deviation to the mean) might be useful
here. For example, a coefficient of 10% or more might be classified as a high degree of
uncertainty.
 Comments on Practical Applications 187

jectured that greater differences in uncertainty among groups A and B implies

greater differences in organizational orientation and greater difficulty in achiev-
ing effective collaboration between the groups. Such groups literally speak
different organizational languages, and, if they are t o relate effectively, require
a third party (project manager) who speaks both languages to translate and
integrate for them. Coordination of two groups both having high uncertainty
can be more direct, possibly handled by committees, while two groups, both
having low uncertainty usually present no special coordination problems.
Thus, it is possible for a project manager t o use the project network, along
with the principles of behavioral science, t o pinpoint possible areas which re-
quire special attention t o coordinate related project activities.

REFERENCES

1. Long, J. M., "Applying PERTICPM to Complex Medical Procedures," Pro-

ceedings of Seminar on Scientific Program Management, Department of
Industrial Engineering, Texas A & M University, June 1967.
2. Davis, Gordon, President, DDR International, Atlanta, Georgia, personal
interview, November 198 1.
3 . Jones, Patricia Cukor, Senior Project Manager, Hanscomb Associates, Inc.,
Atlanta, Georgia, personal interview, December 1981.
4. Galbraith, Jay, Designing Complex Organizations, Addison-Wesley Publish-
ing Co., Reading, Mass., 1973.
5. Kerzner, Harold, Project Management, A Systems Approach to Planning,
Scheduling, and Controlling, Van Nostrand Reinhold Co., New York, 1979.
6 . Martin, Charles C., Project Management: How to Make it Work, AMACOM,
New York, 1976.
7. Youker, Robert, "Organization Alternatives for Project Managers," Man-
agement Review, November 1977.
8. Stuckenbruck, L. C., The Implementation of Project Management, Addison-
Wesley, Reading, Mass., 1981.
9. Benningson, Lawrence, "The Strategy of Running Temporary Projects,"
Innovation, September 197 1, pp. 3 1-41.
10. Cleland, David I., "The Cultural Ambience of the Matrix Organization,"
Management Review, November 198 1.
11. Seiler, J. A,, "Diagnosing Interdepartmental Conflict," Howard Business
Review, September-October 1963.
12. Lawrence, P. R., and J. W. Lorsch, "New Management Job: The Integrator,"
Harvard Business Review, November-December 1967, pp. 142-151.
II I
I
I
ADVANCED TOPICS I
RESOURCE CONSTRAINTS IN
PROJECT SCHEDULING

Resource allocation is probably receiving more attention today than any other
aspect of PERT/CPM. One reason for this is the worldwide inflation spiral since
the oil crisis of 1973-74, which has pushed the cost of resources required for
project execution higher and higher (thus making them relatively more impor-
tant). Another reason is that soaring interest rates have significantly increased
the cost of borrowing money, which directly affects many large plant construc-
tion and capital investment projects. Slippage of such loan-financed projects can
have a devastating impact on project budgets and cost expectations.
Another reason lies in the continued advances in computer technology over
the past two decades which have made more cost-feasible the application of
sophisticated solution procedures that are often necessary when resource limita-
tions enter the project scheduling picture.
The basic PERT/CPM procedures which produce a detailed project schedule
are limited in the sense that resource availabilities are not considered in the
scheduling process. Those basic procedures implicitly assume that available
resources are unlimited and that only technological (i.e., precedence) require-
ments constrain job start/finish times. One consequence of this is that the
schedules produced may not be realistic when resource constraints are con-
sidered. For this reason the basic time-only PERT/CPM forward-backward pass
procedure has been called by some seasoned users, "a feasible procedure for
producing nonfeasible schedules."
It is important to note in discussing the subject of resource constraints here
192 11 I Advanced Topics

that timelcost trade-off procedures are not included. Those procedures typi-
cally assume unlimited resources, are basically different from the procedures
discussed in this chapter, and are covered in the Appendix to Chapter 8. Of
course, alternative project durations and costs could be covered in the context
of simulation of alternative critical resource levels and their associated costs.
Such an approach is possible with the procedures discussed here, but it requires
extensive computational resources.
Resource constraints, while increasingly important, also unfortunately com-
plicate and alter some of the basic notions of PERTICPM. For example, the
longest sequence of activities through the project when resource availabilities
are constrained may not be the same critical path determined by the basic time-
only PERTjCPM approach. Another difference is that with the basic time-only
procedures there is one unique Early Start time (ES) schedule, while under
resource constraints many different ES schedules may exist. To understand
these differences it is necessary t o look at how limited resources affect sched-
ule slack (float).

HOW LIMITED RESOURCES AFFECT SCHEDULE SLACK

Figure 7-1 shows a simple 11-activity network (activity-on-node format) with

activity times indicated beside each node. Figure 7-2(a) shows the all-ES bar
chart schedule for this network. As can be seen, the project duration is 18
weeks, the critical path is the activity (job) sequence A-C-I-J-K, and jobs B, D, E,
F, G, and H all have positive slack.
Now assume that jobs C and G each require the use of a special piece of equip-
ment, such as a hoist crane. But only one crane is available. And assume that
jobs E and F each require a special bulldozer, but only one is available.
The direct result of these resource constraints is that neither jobs C and

V
k Duration, days

Figure 7-1 Illustrative 11-activity network (activity on node format).

 Resource Constraints in Project Schaduling 193

(a) Unlimited resources &\I ---------I

all-ES schedule for
-I
network of Figure
7-1. ------

I
I
Ib) I f jobs C and G
each require a
1 J
&---{
crane, but only one
is available, and jobs
l&l---+ =A
E and F each FI---~
require a bulldozer
but only one is
available.
B
------
I I I I I I I I I I I I 1 1 1 1
I)
5 10 15 18

Figure 7-2 Bar chart schedules for network of Figure 7-1.

G nor jobs E and F can be performed simultaneously as indicated by the ES

time-only schedule. One or the other of the jobs in each pair must be given
priority and each pair must be sequenced so there is no overlap, as shown in
Figure 7-2(b).
Examination of Figure 7-2(b) shows that, when resources for activities C/G
and E/F, respectively, are constrained, the following is apparent:

Activities G and H become critical, with slack reduced to zero.

Activities D, E, and F have their slack reduced significantly.
With activity E given priority over F as shown (an arbitrary choice), the
slack of jobs D and E become dependent upon job F.
No job can start earlier than shown, given the precedence relations and
resource constraints, so this represents an early start schedule. However,
this schedule is not unique (as is the case with unlimited resources), since
job F could have been sequenced before job E in resolving the resource
conflict. In that case, the resulting schedule, though only slightly different
194 11 I Advanced Topics

from the one shown here, would be another ES schedule for the resource-
constrained case.

As this example illustrates, slack can be affected in significant ways when

resources are limited. In general, the following is true:

Resource constraints reduce the total amount of schedule slack.

Slack depends both upon activity precedence relationships and resource
limitations.
The early and late start schedules are typically not unique. This means that
slack values are not unique. These values depend upon the scheduling rules
used for resolving resource conflicts.
The critical path in a resource-constrained schedule may not be the same
continuous chain(s) of activities as occurring in the unlimited resources
schedule. A continuous chain of zero-slack activities may exist, but since
activity start times are constrained by resource availabilities as well as pre-
cedence relations this chain may contain different activities.

MULTIPROJECT SCHEDULING

The impact of resource constraints illustrated by the single-project example

above is magnified in scheduling multiple projects, i.e., situations where several
separate, nominally independent projects are linked together through their de-
pendence upon a pool of common resources. When the pool is tightly con-
strained relative to total concurrent requirements, activity sequencing choices
similar to that illustrated above must be made. However in this situation the
ramifications of the sequencing decision may extend beyond the boundaries of
the project containing the particular activities involved.
Figure 7-3 illustrates a hypothetical 3-project scheduling situation involving
only 3 resource types. To further simplify the example imagine that activities
requiring a resource use only one unit of any one of the three types, and only
one unit of any type is available (e.g., as in a machine-limited workshop). All
activities are shown on their respective resource bar charts at their early start
times.
It is not difficult to visualize the dominolike series of events that might occur
(depending upon activity float, and project fmish times) as a result of delaying
activities to resolve particular resource conflicts. For example, delaying activity
1-3 of project 1 (to resolve the conflict with activity 1-3 of project 2) might
cause the following:

a. delays in successor activities 3-4, 3-5, and 4-5 of project 1;

b. as a result of (a), the creation of additional resource conflicts among activ-
ities requiring resource types 2 and 3 (which must be resolved);
 Resource Conatrainb i n Projact Scheduling 195

Project 1 Proiect 2 Project 3

I
Time
I
Time
L
Time
Resource 1 Resource 2 Resource 3
Figure 7-3 Example of rnultiproject scheduling interactions.

c. as a result of (b), additional delays in projects 2 and 3, and possibly even

project 1 again.

In practice, the schedules for such multiproject problems, involving tens of

projects, dozens of resource types, and thousands of activities, are capable of
development only with the aid of large-scale computers. Some examples of such
computer systems are given in Chapter 11; the point here is that it is the aspect
of resource constraints which elevates the multiproject problem from a relatively
c
simple exercise, when the usual time-only CPM procedures are involved, to a
problem of truly formidable proportions that can require sophisticated compu-
tational routines and powerful computers for solution.

RESOURCE LOADING

One of the main advantages of the network model for project planning is the
ease with which information about resource requirements over the duration of
the project can be generated. The only condition for obtaining this information
is that the resource requirements associated with each project activity shown on
the network be identified separately.
For example, Figure 7-4 shows the same network as Figure 7-1 with man-
power requirements of two different types indicated above each activity. By
utilizing these resource requirements in conjunction with both an early-start
 (3. 1)

Activity no.

k Duration, days

Figum 7-4 Network of Figure 7-1 with resource requirements added.

schedule (such as shown in Figure 7-2(a)) and a late-start schedule (not shown)
the protides of resource usage over time as shown in Figure 7-5 are obtained.
These profdes are commonly called resource loading diagrams. Such diagrams
are extremely important in project management; they highlight the period-by-

12 -
- Resource A
10-
8-
-
4 - I
-----J
@= 2- I
e 1-
I
I
.-z 0 ' " ' I I I I I I I I I I I I I
3
C 5 10 15 18
a, Days
e
Resource B

Days

Figure 7-5 Resource profiles for network of Figure 7-4.

 Resource Constraints in Project Scheduling 197

period resource implications of a particular project schedule and provide the

basis for improved scheduling decisions.
Resource Loading diagrams are probably more widely used than any other
resource analysis technique in project management. The technique is simple,
provides a useful visual picture, and does not always require a computer (except
for large-sized networks). In many cases the resource loading data are more im-
portant as rough planning guides rather than as precise goals to be obtained,be-
cause of schedule uncertainty. One major pharmaceutical firm, for example,
obtains resource loading- information from all CPM schedules for research and
development projects, to provide resource feasibility checks on the amounts of
high-skilled
- medical and technical research personnel required. Such checks are
particularly useful in the case of multiple on-going projects, which each require
varying amounts of resources that must be shared by all projects. I

Cumulative Resource Requirements I

I

Figure 7-6(a) shows the period-by-period total requirements of units of resource

A of network Figure 7-4, for both the early-start (ES) and late-start (LS) sched-
ules. The period totals were used in constructing the resource loading diagram
shown in Figure 7-5. Also shown in Figure 7-6(a) are the cumulative require-
ments for each period. These latter data can be usefully displayed in the form
of cumulative resource requirement curves as shown in Figure 7-6(b).
The cumulative requirements curves can be used to develop project sched-
uling information both before and during project implementation. For example,
as time progresses after the project has started, the cumulative resources required
should lie within the closed area bounded by the ES and LS cumulative curves.
If actual cumulative resources fall under the LS curve the project is either behind
schedule or the resource requirements were overestimated. Conversely, if they
exceed the ES curve the project is either ahead of schedule or the resource re-
quirements were underestimated.

Use in Resource Planning

Another important use of the cumulative curves is in preliminary resource alloca-

tion planning. The magnitude of total cumulative requirements and the slope of
the average requirements line drawn to this end point can be used to develop
rough indicators of resource constraint "criticality" and of the likelihood of
project delay beyond the initial forward-pass determined critical path duration.
For example, lime I in Figure 7-6(b) indicates the average daily requirements for
resource A as:

11 1 total units
avg. daily requirement = = 6.2 unitslday
18 days
198 11 I Advanced Topics

ES Schedule L S Schedule

Total Cum. Total Cum.

Period Units Units Units Units

1 5 5 3 3
2 5 10 3 6
3 5 15 3 9
4 9 24 3 12
5 9 33 5 17
6 9 42 5 22
7 9 51 5 27
8 9 60 5 32
9 6 66 9 41
10 5 76 10 51
11 5 76 9 60
12 5 81 9 69
13 5 86 9 78
14 6 92 6 84
15 4 96 6 90
16 5 101 7 97
17 5 106 7 104
18 5 11 1 7 111

Figure 7-6(s) Early and late start schedule requirements of resource A, network of Figure
7-4.

Days
Figure 7-6(b) Cumulative resource requirements of resource A, network of Figure 7-4.
 Resource Constraints in Project Scheduling 199

Suppose resource A is available at a maximum level of 7 unitslday. In this case

a total of 126 units could be expended over the 18-day critical path duration,
which is considerably more than the 111 units required over that period. Thus,
there is unlikely t o be a project delay beyond the l&day duration because of
constraints on Resource A. This conclusion can also be drawn from the ratio
of resources required to available, which is a rough measure of resource "tight-
ness," or criticality. That is,
avg. daily units req'd. - 6.2
criticality index = - 0.88
max. am't. avail. daily 7.0
The situation above can be contrasted to the case where only 6 units of re-
source A are available daily. In this case
62
criticality index = - = 1.03,
6.0
In 18 days a total of only 108 units of resource will have been expended (or "in
place"), leaving some work unfinished and thus requiring an extension of the
project beyond 18 days.
In general, higher values of the resource criticality index calculated here are
associated with the most critical (i.e., most tightly constrained) resources. Very
little research has been conducted in this area, but what has been done7 suggests
that, while exceptions exist, values significantly below 1.0 typically are asso-
ciated with nonconstraining resources, while values around and above 1.0 indi-
cate that project delays beyond the original critical path duration will be
encountered.
Figure 7-7 shows another type of situation which can exist when using cumu-
lative resource requirement curves. This shows ES and LS cumulative require-
ments curves for one type resource (cubic yards of concrete poured) for a
hypothetical project. From the position of the ES and LS curves relative t o the
project start date, it can be seen that concrete is not required during the very
beginning or end periods of project duration. It is assumed here that total proj-
ect duration has been estimated from the forward pass computations to be 200
working days, with concrete required only between days 65 and 185 (ESsched-
ule). A total of 1,000,000 cu. yds. is assumed to be needed and a maximum of
6,000 cu. yds./day is available.
Line I shown in Figure 7-7, constructed as in the previous example from
project start to the point of total cumulative resource requirements at project
completion, gives the average daily requirement for concrete as:
1,000,000 cu. yds.
avg. daily requirement over project duration =
200 days
= 5000 cu. yds./day
200 11 I Advanced Topics

5 1,000.000 -
3

4.rn

C
C
8
-
+
0
w

E,
C
w
.-
-
4-

2 /
/
/
5
U /
/

0 65 165 200
t
Start
t
Project
time duration

Figure 7-7 Cumulative requirements curves for hypothetical concrete pouring example.

The use of this average for estimating resource criticality, as in the previous
example, will lead to a conclusion that concrete is not a constraining resource.
But this is misleading; this average is clearly a lower level of usage than is really
required, because the usage is projected over the entire project duration instead
of the shorter period of actual need. A better measure of the average require-
ment is given by Line 11, which is drawn from the start of the ES curve to the
total cumulative requirements level at the end of the ES curve. The average
requirement given by the slope of this line is calculated as:
1,000,000 cu. yds.
avg. daily requirement over actual use period =
(185 - 65) days
= 8333 cu. yds./day.

If this average is used as before for estimating resource criticality a dif-

ferent conclusion is drawn: The criticality index is geater than 1.0, only
720,000 cu. yds. can be poured over the indicated period of 120 days, and
project delays beyond day number 200 will occur. The project can be suc-
cessfully completed within 200 days only if concrete pouring can start suffi-
ciently earlier than day 65 (e.g., by changing network logic, etc.) to accomplish
installation of the 1,000,000 cu. yds. by day 200, and other resources do not
become more critical than concrete.
Other situations similar to F i r e 7-7 include cases of intermittent or irregular
requirement of the resource. For example, Figure 7-8 illustrates a case where a
relatively small amount of some resource is required shortly after project start,
followed by a period of time of zero requirements, with significant requirements
in succeeding periods until shortly before project end (ES schedule).
 Resource Constraints in Project Scheduling 201

Praject end
start Days
Figure 7.8 Illustration of resource lower bound calculation from cumulative resource
requirement curves.

In cases such' as this an estimate of the minimum level requirements of the

resource is obtained by construction of the minimum-slope line which can be
drawn within the ES, LS "envelope" of cumulative requirements. Drawn tan-
gent to the envelope in Figure 7-8 at points (a) and (b), it gives the following
lower-bound estimate:
L2 - L 1 resource unitslday
lower bound estimate of avg. daily requirement = -
T2 - T I
The estimate of the average calculated in this fashion can then be used in calcu-
lating the index of resource criticality, as shown earlier.
The use of the cumulative resource requirements curves as described generally
in this section provides an approach for estimating the impact of resource con-
straints, either in advance of or in the absence of more sophisticated procedures.
The procedure could also be used, if desired, as a basis for setting activity prior-
ities in project scheduling. To do this, all resources required by the project
could first be ranked by criticality; this listing could then be used to rank the
activities, putting those that require the most critical resource(s) at the top of
the list. Then the activities could be scheduled in the order listed.
All good computer programs for CPM analysis today will produce resource
loading information (ES, LS daily profiies and cumulative requirements) in the
form of tables or figures even though they may lack the more sophisticated
capability of automatic schedule adjustment to meet desired resource loading
requirements. The latter capability falls into the categories of resource leveling
or scheduling under fmed resource constraints, which are discussed in the next
section.
202 11 I Advanced Topics

OVERVIEW OF BASIC SCHEDULING PROCEDURES

Scheduling procedures for dealing with resource constraints can be roughly

divided into 2 major groups, according to the problem addressed: (1) resource
leveling, and (2) fixed-limits resource scheduling.
1 . ResourceLeveling. This type of problem occurs when sufficient total
resources are available, and the project must be completed by a specified due
date, but it is desirable or necessary to reduce the amount of variability in the
pattern of resource usage over the project duration. This type of situation
occurs frequently, for example, in the construction industry, where the cost of
hiring and laying off of personnel or physical resources can be substantial. Thus,
the objective is to level, as much as possible, the demand for each specific re-
source during the life of the project. Project duration is not allowed to increase
in this case.
2. Fixed Resource Limits Scheduling. This category of problem, which is
much more common, arises when there are definite limitations on the amount
of resources available to carry out the project (or projects) under consideration.
The scheduling objective in this case is to meet project due dates insofar as
possible, subject to the fvred limits on resource availability. Thus, project dura-
tion may increase beyond the initial duration determined by the usual "time-
only" CPM calculations. The scheduling objective is equivalent to minimizing
the duration of the project (or projects) being scheduled, subject to stated con-
straints on available resources. This category of problem can be further sub-
divided according to whether the fuced limits on resource availabilities are
constant at some level or allowed to vary over activity or project duration. A
further useful subdivision is possible according to whether approximate, rule-of-
thumb procedures, or mathematically exact procedures are employed. Davis6
presents a general classification scheme employing these and other ideas.
The basic general approach followed in both resource leveling and fixed re-
source limits scheduling is similar: Set activity priorities according to some
criterion and then schedule activities in the order determined, as soon as their
predecessors are completed and adequate resources are available. Because of
this common general approach, it is useful to understand somewhat better the
nature of the criteria employed before proceeding with a discussion of the dif-
ferent problems.

USE OF HEURISTICS I N SCHEDULING

The task of scheduling a set of project activities such that both precedence rela-
tionships and constraints on resources are satisfied is not an easy one, even for
projects of only modest size. The difficulty is increased if simultaneously some
objective such as minimum project duration or minimum total cost is sought.
 Resource Constraints in Project Scheduling 203

The l i i t e d resource project scheduling problem falls into a category of math-

ematical problems known as cornbinatoriaI problems. This is because, for any
given problem, a very large number of possible combinations of activity start
times exist, with each combination representing a different project schedule.
The number of combinations is extremely large, even for fairly small problems
of 20 or 30 activities and increases rapidly with an increase in the number of
activities. In fact, the number is typically so large for realistic-sized problems
as to prohibit enumeration of all alternatives, even with the aid of a computer.
Analytical methods such as mathematical programming have not proven very
successful on these combinatorial problems. Instead, various heuristic-based pro-
cedures have been developed.
Simply stated, a heuristic is a rule of thumb-a simple, easy to use guide or
aid used in problem-solving situations, t o reduce the amount of effort required
in coming up with a solution. Heuristics may not always produce the best
solution in every case, but their usefulness in finding good solutions with a
minimum of effort is well-known, based on experience and research studies.
Relatively simple heuristics such as "shortest job first," or "minimum slack
first" are effective aids in establishing activity priorities on many types of
limited-resource scheduling problems. An example application of such simple
heuristic rules will be illustrated here on both categories of problem listed above.
More complicated heuristic procedures-typically consisting of combinations of
heuristics and modifying rules-will be discussed later.

PARALLEL vs. SERIAL METHODS

There are two general methods of applying heuristics in project resource alloca-
tion problems. A serial scheduling procedure is one in which all activities of the
project are ranked in order of priority as a single group, using some heuristic,
and then scheduled one at a time (i.e., serially). Activities that cannot be started
at their early start time are progressively delayed until sufficient resources are
available.
In parallel scheduling, all activities starting in a given time period are ranked as
a group in order of priority and resources allocated according to this priority as
long as available. When an activity cannot be scheduled in a given time period
for lack of resources, it is delayed until the next time period. At each successive
time period a new rank-ordering of all eligible activities is made and the process
continued until all activities have been scheduled.
Even though the parallel method requires more computer time to reorder the
eligible activities at each time period, it appears to be the more widely used of
the two methods, being employed in a number of commercially-available com-
puter programs for project scheduling. The only published research comparing
204 11 I Advanced Topics

the effectiveness of the two approaches in limited resource scheduling indicated

that the serial approach could produce shorter-duration schedules for some
categories of networks, but that there were also disadvantages with this p r e
cedure in terms of the special scheduling conditions (such as activity splitting)
which could be handled.''

RESOURCE LEVELING

As noted earlier, in many project scheduling situations it is the p a t t m of re-

source usage, such as frequent changes in the amount of a particular manpower
category required, or peaks of resource demand at undesirable times, that must
be improved. The concentration of highest levels of demand for resource type
R1 in the earlier schedule periods, as seen in Figure 7-5, might be such an ex-
ample. Resource-leveling techniques are useful in situations such as this; they
provide a means of distributing resource usage over time to minimize the period-
by-period variations in manpower, equipment, or money expended.
The essential idea of resource leveling centers about the rescheduling of ac-
tivities within the limits of available slack (float) to achieve better distribution
of resource usage. The slack available in each activity is determined from the
basic scheduling computations, without consideration of resource requirements
or availabilities. Then, during the rescheduling, or "juggling" of activities t o
smooth resources, project duration is not allowed to increase.
While the essential ideas of resource-leveling are easy to grasp, implementation
in practice can pose difficulties. For example, what activity schedule will mini-
mize resource costs for the network of Figure 7-4 given that project duration is
not to exceed 18 days? Questions such as this are not always easy to answer just
by looking at resource profiles such as shown in Figure 7-5. Instead, some sys-
tematic procedure for arriving at the best (i.e., most level resource utilization)
schedule is needed which provides an easily calculated measure of effectiveness.
The first edition of this text described a systematic procedure developed by
Burgess2 for leveling resources. This method utilized a simple measure of effec-
tiveness given by the sum of the squares of the resource requirements for each
"day" Cperiod) in the project schedule. It is easy to show that, while the sum of
the daily resource requirements over the project duration is constant for all com-
plete schedules, the sum of the squares of the daily requirements decreases as the
peaks and valleys are leveled. Further, this sum reaches a minimum for a sched-
ule that is level (or as level as can be obtained) for the project in question.
The Burgess procedure which utilizes the above measure of effectiveness is
simple and can be utilized under a variety of problem assumptions (e.g., activ-
ity splitting). The basic method, which is representative of a general class of
resouce-leveling procedures, is explained in the following section.
 Rwurcs Constmintsin Project Scheduling 205

Burgess Leveling Procedure

Step I. List the project activities in order of precedence by arranging the

arrow head or node numbers in ascending order. When using activity-on-arrow
format and two or more activities have the same head number, list them so that
the arrow tail numbers are also in ascending order. (This assumes that the net-
work events are numbered so that activity tail numbers are always less than the
head numbers.) The procedure given in Appendix 4-1 includes a routine t o ac-
complish this. Add to this listing the duration, early start, and slack values for
each activity.
Step 2. Starting with the last activity (the one at the bottom of the diagram),
schedule it period by period t o give the lowest sum of squares of resource re-
quirements for each time unit. If more than one schedule gives the same total
sum of squares, then schedule the activity as late as possible to get as much slack
as possible in all preceding activities.
Step 3. Holdimg the last activity foted, repeat Step 2 on the next to the last
activity in the network, taking advantage of any slack that may have been made
available to it by the rescheduling in Step 2.
Step 4. Continue Step 3 until the first activity in the l i t has been considered;
this completes the first rescheduling cycle.
Step 5. Carry out additional rescheduling cycles by repeating Steps 2 though
4 until no further reduction in the total sum of squares of resource requirements
is possible, noting that only movement of an activity to the right (schedule later)
is permissible under this scheme.
Step 6. If this resource is particularly critical, repeat Steps 1 through 5 on a
different ordering of the activities, which, of course, must still list the activities
in order of precedence.
Step Z Choose the best schedule of those obtained in Steps 5 and 6.
Step 8. Make final adjustments to the schedule chosen in Step 7, taking into
account factors not considered in the basic scheduling procedure.
If Steps 2-4 of the Burgess procedure are applied to the sample network of
Figure 7-4, the sequence of activity shifts, total units of each resource type r e
quired and calculated sum of squares at each major step are as shown in Figure
7-9, along with the final activity positions. Figures 7-lqa) and (b) show, respee
tively, the sum of squares and resource profiles before and after application of
the Burgess procedure. As can be seen, application of the approach produces
both a lower sum of the squares and generally smoother profiles for both re-
sources, although the peak level of resource A has actually increased slightly
compared to its initial level. Resource A can be held t o a peak period require-
ment of 8 units through Step 8 of the procedure ("final adjustments to the
schedule"; here by changing slightly the positions of activities E and G). How-
 Sequence o f Major Moves
Tim N
(Each scriviry shifted in lporiod
Acririry 1 2 3 4 5 6 7 8 9 10 11 12 73 14 15 16 77 18 incmmenn for tDfalrhownl 8
A' 3A 3A 3A
28 2% 28
(1) Delay H one period
ZR' = 1132
---
121 Delay F two periods
Z R 2 = 1118

-
(31 Delay Gone per~od
Z R Z 1116

14) Delay E five periodr

ZR' = 1112

151 Delay D rwo periods

m1= 1062

K.

.critics1 activity

Levelof 5 5 5 9 9 9 9 9 6 5 5 5 5 6 9 5 5 5 ~ ~ ' = 7 4 733

7 727 723 715
Resource A 5 5 8 8 7 7 3 7 7 7 6
Assigned

Lwalof 6 6 6 8 8 3 3 3 2 4 4 4 4 3 1 4 4 4 ZB'= 385 -

- 385 -
389 -
389 -
397
Rerource 8 5 5 7 7 3 2 3 Total 1132 1118 1116 1112 1062
Asigned 1 5 3
4 4

Fiaure 7.9 Results of aDDlication of Burgess procedure to network of Fimre 7-4.

 Resource Constraints in Project Scheduling 207 i

A l l Activities A l l Activities After Leveling

at ES Time at L S Time via Burgess

Max. Res. Sum o f Max. Res Sum of Max. Rea Sum of

Resoume Level Squares Level Squares Level Squares

A 9 743 10 779 10 715

B 8 389
- 8 387
- 6 347
-
1132 1166 1062

Figure 7-lola) Comparison of Burgess leveling criterion for neiwork of Figure 7-4.

1
: :
'2
"
[---J-t-c n
Before leveling
/IESschedule)
' (After
'

leveling

L-l
L-----A %---
L-J

Days
g
'5
m 9 -
10-

E! 8- ,----t
m
s
f
7 -
6.
I I
E! 5 -
C
, 4-
3 -
I
S 2-
1 - L-1
1 1 1 1 1 1 1 1 1 1 \ 1 1 1 1 1 1 /
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Days
Figure 7-10lbl Resource profiles before and after application of Burgress procedure.

ever, this can be done only at the expense of increased variability in the level of
resource B, with a corresponding increase in the total sum of squares.
Burgess' ideas are utilized in a recent paper by Woodworth and who
present a multiproject approach in which the sum of squares smoothing criterion
is applied to each resource sequentially, across all projects. Although the authors
do not make it explicit, the procedure would presumably be applied to resources
in some descending order of criticality (i.e., most critical resource first, etc.).
208 11 f Advanced Topics

FIXED RESOURCE LIMITS SCHEDULING

The process of resource leveling as illustrated above will smooth the profiles
of resource demand to the extent allowed by available activity float. But this
process does not always produce satisfactory schedules if the amounts of avail-
able resources are tightly constrained. The final profiles of resource usage in
Figure 7-10(b), for example, show levels that might still exceed the maximum
amount of resources available. What would happen, for instance, if the amounts
of Resource A and B were limited to, say, 6 and 7 units per day, respectively?
The answer to this question requires the use of the second category of pro-
cedures mentioned previously, called fixed resource limits scheduling. Also
often called constrained-resource scheduling, or limited resource allocation,
these are techniques designed to produce schedules that will not require more
resources than are available in any given period, with project durations which are
increased beyond the original critical path length as little as possible.
An example flow chart of this type of scheduling procedure is shown in
Figure 7-11. The procedure will be applied to the same sample network of

7START

Calculate initial early start (ES) and late start (LS) time for each activity
in the project, and set time now equal to 1, i.e., T = 1.

f
Determine the initial eligible activity set (EAS), i.e., those activities with
all predecessor activities scheduled.
1
Fmm among the members of the current EAS, determine the ordered
-scheduling set (OSS) of activities, i.e., activities with ES 5 T , ordered
according to LS with smallest values first and within this characteristic,
according to least activity duration first.
t
Consider the activities in OSS in the order listed and schedule those
activities for which sufficient resources are available for the duration of
the activity. A s activities are scheduled update the level of resources
available, and update the members of EAS.
t
Have all activities been scheduled,
i.e., is EAS an empty set?
t
No
yes -m
Set T new = T old +1 and compute new ES times for the updated
EAS. I
~ - - - -- -.

Figure 7-11 Basic constrained-resourcescheduling procedure.

 Rerource Constraints in Project Scheduling 209

Figure 7-4 that we have been using, but with the amounts of Resource types A
and B available limited to 6 and 7 units per period, respectively.
The procedure involves a parallel approach (stepping through time, period-by-
period). First a set of activities is defined whose predecessors are all scheduled;
this is called the Eligible Activity Set (EAS). Then, only the activities in the
EAS with ES less than or equal to some time value T are considered. These
activities are ordered with least slack first and, within this criterion, with short-
est duration first. This ordered list of activities is referred to as the Ordered
Scheduling Set (OSS).
The bookkeeping for this scheduling procedure can be simplified by ordering
the activities in the OSS according to their late start time (LS) instead of by
slack. The minimum-LS ordering is identical t o the ordering obtained by using
minimum slack, as shown in Reference 4. However, an advantage is that the LS
values do not change from time period t o time period, whereas slack values con-
tinuously decrease for an activity that is ready to be scheduled in a given period
but is not scheduled.
The complete schedule shown in Figure 7-12 can be obtained by continuing
ACTIVITY Time
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22

L ~ Y ~ I O ~ A ~ ~ ~ $ B P B B B ~ ~ B G
u n a s r i g n e d $ $ $ ~ 4 a 3 P ~ 2 2 $ 3 3 ~ 4 4 4 14 11
1 1 1 1 1 0 0 0 0 i x t 1 o o
0 0 0
level of^ 1 f f I f f f f j # I I i f ? # i i f I f i
" n a a i g n e d % 5 5 $ # $ $ d B 4 4 $ $ $ d p l $ 5 5 3 3 3
1 1 1 2 2 5 5 5 5 4 4 4 3 4 4
Figure 7-12 Application of basic scheduling procedure to network of Figure 7-4.
210 11 I Advanced Topics

the above procedure as indicated in Figure 7-13. The final schedule shows that
project completion is delayed 4 days past the early finish time of 18 days com-
puted for this project without regard to limits on available resources.
The procedure illustrated by this example represents a potentially powerful
general approach to a broad class of scheduling problems. The procedure can be
applied to multiple project problems as well as single projects. For the multi-
project case, project start and end dates for each project are necessary. This in-
formation is used to compute total activity slack (float), or the equivalent late
start times, to permit determination of the order in which various activities
are considered for scheduling. It should also be emphasized that the number
of resources being considered is not limited by this procedure but only by the
capacity of the computing system being used. The basic approach illustrated
by this simple example is, in fact, essentially how the majority of heuristic-
based constrained-resource scheduling systems operate, including some large
commercially-offered computer programs. However, the basic steps indicated
here are often embellished considerably to permit more realistic simulation of
the numerous variations followed in the actual practice of scheduling project
activities. Some examples of these embellishments will be given later.

Relative Effectiveness of Heuristic Rules

If a heuristic other than minimum slack had been used for ordering the activities
in the OSS,a different schedule might have been produced for the example net-
work. The min-slack heuristic was used here because it has been found, in sev-
eral studies, to generally produce the best results. For example, a 1975 study4
comparing the 8 rules shown in Table 7-1 showed that the minimum slack rule
generally outperformed the 7 other rules, as shown in Figure 7-14. These results
were based on a sample of small (i.e., 27-activity) multiresource problems for
which the optimal solution, in terms of minimum schedule duration, could be
calculated. A different study which involved larger networks fronl actual prac-
tice ranging in size up to 180 activities showed that either a Late-Start (equiva-
lent to min-slack) or Late-Finish heuristic was most effective for both parallel or
serial allocation procedures.10 However, that study also concluded that the
choice of allocation procedure (i.e., parallel or serial) was of greater significance
in some cases than the choice of scheduling heuristic.
Although individual studies such as the two cited above have indicated the
general best-effectiveness of a particular heuristic, or type of heuristic, it must be
emphasized that all such studies have also shown that no one heuristic-or com-
bination of heuristics-always produces the best results on every problem. This is
perhaps the greatest disadvantage of heuristics: Rules that perform well on one
problem may perform poorly on another, and vice-versa. In practice, even with
more sophisticated procedures, it is not possible to absolutely guarantee in ad-
T- 1 T = 12
EAS: A B EAS: I E F
ES. 1 1 ES: 12 12 12
LS: 1 5 Ls: 10 11 11
OSS: A 8 DSS: 1 E F
Schedule A i n periods 1-3 and remove i t Schedule I i n periods 12-15 and remove it
from EAS. from EAS.
Schedule B i n periods 1-5 and remove i t Schedule E and F i n periods 12-14 and
from EAS. remove them from EAS.
Add C. D, and G t o EAS. Add J and H t o EAS.

T=2 T = 13
EAS: C D G EAS: J H
ES: 4 4 6 ES: 16 15
LS: 4 9 10 LS: 14 14
0%: None-No activities can be rched- DSS: None-No activities cmn be sched-
"led. uled.
-
S~ncem8nlmum ES 4, r k l p T = 3 and go t o
T=4.
Since minimum ES = 15, skip T = 14 and
gotoT=15.

T =4 T = 15
EAS: C 0 G EAS: J H
ES. 4 4 .6 ES: 16 15
LS: 4 9 10 LS: 14 14
OSS: C D OSS: n
Schedule C i n periods 4-9 and remove i t Schedule H i n periods 15-19 and remove it
from EAS. from EAS.
A D D I t o EAS. T-16

t5: 10
EAS. D G I
LS: 14
ES:
LS:
ncc.
5
9
n
6
.- --
10
05s: J

T =6 T = 17
EAS: D G I EAS: K
ES: 6 6 9 ES: 18
Ls: 9 10 10 LS: 16
OSS: D G 055: Nona-no sctiviries can be sched-
Schedule G i n priods 6-9 and remove it uled.
from EAS.
Since D can be scheduled no sooner than T=18
perlod 10 because of resource con- EAS: K
straints, t o go T = 10 = ES of I. ES: 18
LS: 16
T = 10 OSS: K no activities can be sched-
EAS: D I uled since K can be
ES: 10 10 scheduled n o sooner
LS: 9 10 than period 20 because
oss: D I of resource constraints,
Schedule D i n perrods 10-11 and remove it go t o T = 20.
from EAS.
T=20
Add E and F t o EAS.
EAS: K
T = 11 ES: 20
EAS: I E F LS: 16
ES: 11 12 12 OSS: K
Ls: 10 11 11 Schedule K i n periods 20-22 and remove it
OSS: I from EAS.
No actbvitier can be scheduled. All activities have been schaduled. STOP.

Figure 7-13 Details of bssic scheduling prwadure applied to network of Figure 7-4.
212 11I Advanced Topics

Table 7-1. Some Heuristics Used in Constrained-Resource Scheduling.

Heuristic Scheduling Rules Evaluated

Rule Notation Operathg Features

Minimum Activity Slack MINSLK Schedules first those activities with lowest
activity slack time (total float).
Minimum Late Finish Time LFT Schedules first those activities with the
earliest values of late finish time.
Resource Scheduling Method RSM Priority index calculated on basis of pair-
wise comparison of activity early finish
and late start times. Gives priority to ac-
tivities roughly in order of increasing late
finish time.
Greatest Resource Demand GRD Schedules first thoseactivities with greatest
resource demand in order to complete po-
tential bottleneck activities.
Greatest Resource Utilization GRU Gives priority t o that group of activities
which results in the minimum amount of
idle resources in each scheduling interval.
Involves an integer linear programming
logarithm.
Shortest Imminent Operations SIO Schedules first those activities with short-
est durations in an attempt to complete
the greatest number of activities within a
given time-span.
Most Jobs Possible MJP Gives priority to the largest possible group
of jobs which can be scheduled in an in-
terval. Involves an integer linear program-
ming logarithm.
Random Activitv Selection RAN Priority given t o jots selected at random,
subject to resource availability limits.
- - - - -- -

Note: The MINSLK rule has been shown equivalent t o a minimum-late-start-time rule.

vance which particular heuristic-or combination of heuristics-will produce

best results for a given problem.
In spite of this drawback, heuristic procedures are used very widely in prac-
tice. The schedules produced by these procedures may not be the theoretically
best possible, but they are usually good enough to use for planning purposes in
view of the uncertainties typically associated with activity durations and re-
source constraints and requirements. Furthermore, some very powerful com-
puter-based solution procedures incorporating a variety of imaginative heuristics
have been developed which will produce schedules for large, complex projects
under a variety of special assumptions such as job splitting, "crashing," stretch-
ing out of jobs by varying the rate of resource application, etc. General informa-
 Resource Constraints i n Project Scheduling 213

Percentage of
problems for
which optimum

GRD
RAN
GRU
MJP
SIO

(a) Percent of problems for which optimal minimum duration solution was obtained.
(Optimal vs. nonoptimal totals include problems in which one ormore heuristic sequencing
rules produced a minimum duration solution.)

Performed relatively better

than a random rule

RSM 1 6.8% 1 J
RAN

GRD
Performed relatively worse
than a random rule
SIO

MJP 16.1% 1 1
(b) Percent increase above optimal duration.

Figure 7-14 Heuristic rules result^.^

tion about a number of such commercially-availablecomputer programs is given

in Chapter 11. Many other such programs have been developed by individual
companies for proprietary use. The details of these approaches are not generally
available for publication.* However one program, representative of this class of
comprehensive heuristic procedures, that has been described in the open litera-
ture is described briefly below.

Wiest's SPAR-1 Model

One of the more comprehensive heuristic scheduling procedures developed by

an independent researcher is called SPAR-I (for Scheduling Program for Allocat-

*These programs vary widely with regard to approaches used in scheduling under resource
constraints Some, such as the McAuto "MSCS" package, allow the user t o choose from
among several available heuristic rules or even create his own. Others utilize one or more
built-in heuristic rules without allowing any user choice.
214 11 I Advanced Topica

ing Resources). Developed in the mid-1960s by J. D. W i e ~ t ? it ~ has

'~~ served as
a model for many other subsequent procedures. A flow diagram of the pro-
cedure is given in Figure 7-15.
The SPAR-1 model is similar to the basic procedure diagrammed earlier in
Figure 7-11, in that a parallel scheduling approach is followed and activities are
sorted for scheduling within each time period in order of their total slack. How-
ever, the SPAR-l model includes a number of useful embellishments, as will be
seen in the following description. One unusual feature not apparent from the
flow diagram is a probability-based selection procedure for determining whether
a job from the list of jobs eligible for scheduling in each period is actually sched-
uled. Under this approach the f i s t job on the list is not automatically selected
for scheduling, but is selected with some probability less than 100%. Jobs from
the top of the list which are not selected are moved to the bottom of the list.
Eventually all jobs in the list will be scheduled, but the sequence of scheduling
will vary randomly, and hence the project schedule will vary on successive runs
of the model, even though inputs are exactly the same. This feature allows a
number of different schedules to be generated and the best one selected accord-
ing to a specific criterion.
In the SPAR-1 model each job has a normal, maximum, and minimum level
of resources which can be assigned to it. Since the total man-days required for
job performance is assumed to be constant irrespective of resource loading, job
duration can be normal, "crashed" (shortened), or "stretched" (increased),
according t o whether the level of resources assigned is normal, maximum, or
minimum. Resource allocation is done automatically by the program, according
to whether the job is critical or noncritical.
Referring to Figure 7-15, it can be seen that the SPAR-1 model iterates day
by day starting with day d = I. Jobs with ES = 1 are selected and ordered ac-
cording to their slack (the "augment critical job routine" actually comes into
play later and will be explained below). If the job to be scheduled is not crit-
ical, attempts are made to schedule it with a normal level of resources ("normal
crew size"). If this is not possible, the minimum crew size is considered before
the job is postponed until day d + 1. If the job to be scheduled is critical a dif-
ferent procedure is followed. First, if sufficient resources are available the job
is scheduled at its maximum allocation; that is, it is crashed. If this is not pos-
sible a normal crew size is considered. If neither of these attempts is successful
two special subroutines are tried. In the f i s t of these, efforts are made to bor-
row needed resources from currently active jobs. Resources are borrowed from
a job only when the resultant "stretching" of the job will not delay the entire
project.
The second subroutine that may be tried for critical jobs reschedules active
jobs to a later start date. That is, jobs that can be postponed without delaying
the project are delayed for scheduling until a later period; the resources thus
 START

Read job data

STOP

Search routine for

Is c o n of schedule< Calculate c o n
b e n previous schedule?

2 max. crew size?

2rnin. crew size?

active job routine

Postpone early
start of job j
today;+ 1 Successful
Decrease men available
by crew size scheduled

Reschedule active

..

Add-on unassigned men

routine

NO
Bookkeeping routine:
i=i+ 1 recalculate tentative
schedule for remaining jobs

Figure 7-15 Flow diagram f o r SPAR-1 scheduling program.

216 11 I Advanced Topics

freed up are available for assignment to the critical job. This reschedule routine
has essentially the same effect as a "look ahead" feature might have. However,
instead of actually looking ahead to predict the needs of critical jobs (which
would be difficult in the limited resources cases, since activities are not always
scheduled at their ES date), the model changes previous decisions as it moves
along from day to day by "taking back" resources that were assigned to jobs
in earlier periods which do not need them as much as the critical jobs.
If neither of the special subroutines (borrow resources or reschedule active
jobs) is successful, a minimum resource level schedule is attempted before the
start of the critical job is postponed. Then, after all possible jobs are scheduled
on day d , the add-on unassigned resources subroutine is called. In this procedure
a list of unused resources is compiled. A list of active jobs requiring these re-
sources that are not at their maximum allocation levels is also compiled, in
ascending order of slack. The assigned levels of these jobs is then increased,
until either jobs or resources is exhausted.
Whenever the program iterates to a new day (i.e., day d + I), the augment
critical jobs subroutine is tried. With this feature jobs previously scheduled and
still active are examined. If any of these is critical and has a resource level less
than its maximum-and if resources are still available-that job's resource alloca-
tion is increased as much as possible up to the maximum.
As the program moves from day to day it also updates all critical path data,
i.e., early and late start and finish times, and job slack. The importance of this
step is apparent, since a job's criticality may be changed through actions taken
by the model. For example, when resources are increased to a critical job the
resultant duration decrease might cause it to gain positive slack and become
noncritical.
When all jobs have been scheduled over all days, the cost of the schedule is
computed, including daily overhead expenses, due date bonuses or penalties,
and resource costs based on the assumed available levels. Depending upon the
cost of the schedule compared with previous schedules, the procedure is either
terminated or a new set of available resources is tested.
SPAR-1 is able to accommodate projects of several thousand jobs, and several
dozen resource types, under a variety of special assumptions, such as job split-
ting, shift or nonshift scheduling, variable resource limits, and single or multiple
projects. The results obtained from SPAR-1 on small tests problems have shown
that in some cases it gave nearly optimal results. However, no heuristic-based
program, irrespective of its complexity, can guarantee to yield an optimal proj-
ect schedule, or even consistently best results from problem to problem, as
noted earlier.
Analytic procedures, such as linear programming, can yield optimal solutions
consistently, but are restricted to relatively small projects with few special con-
ditions. For this reason, and in spite of their limitations, heuristic procedures
 Resource Constraints in Project Scheduling 217

appear to offer, for the foreseeable future, the best approach for scheduling
realistic-sized projects under conditions of limited resources.

OPTIMIZATION PROCEDURES

Another major category of constrained-resource scheduling procedures-those

for producing optimal solutions-has been characterized by relatively less prac-
tical progress than the heuristic methods. The optimization procedures which
have been developed can be divided into two subcategories:

1. procedures based on linear programming (LP);

2. procedures based on enumerative and other mathematical techniques.

Linear programming (LF') was first proposed as a method for solving the con-
strained-resource scheduling problem in the early 1960s. Numerous formulations
of the problem were proposed but found to be impractical except for solving
small problems of only a few activities. Wiest described the use of those early LP
formulations on the constrained-resource problem as "akin to using a bulldozer
to move a pebble."20 Wiest showed that a 55-activity network with four re-
source types would require more than 6,000 equations and 1,600 variables.
Later researchers improved on these early formulations and solved somewhat
larger problems. In recent years, with the increasing capability and decreasing
cost of computers, some researchers have once again begun to investigate the
use of linear programming procedures, either alone or in combination with
other approaches. An example of the latter type of approach is that of Patter-
son and Huber,14 who combine a minimum bounding procedure with integer LP
to reduce the computation time required in arriving at a minimum project dura-
tion. Essentially, their approach involves starting the optimization procedure off
with a "good" lower bound solution, to reduce the domain of possible solutions
over which the LP algorithm must search. The "minimum bound" they use to
initiate the search procedure is simply an estimate of the minimum project dura-
tion implied by the tightest resource constraint, i.e., if the cumulative resource
requirement of that resource is 80 units, and if 8 units per day are the maximum
available, then the minimum project duration = 8018 = 10 days. They tested this
approach on a series of small problems ranging in size up to about 20 activities
and 3 different resource types, with encouraging results, compared to those of
other LP approaches.
Another recent approach is that of ~ a l b 0 t . I ~He uses both integer linear
programming and implicit enumeration in formulating and solving resource-
constrained project scheduling problems in which job performance time is a
function of resource allocation, and total cost is used as an objective function.
His reported computational experience indicates that the methodology can
218 11 I Advanced Topler

provide optimal solutions to small problems and heuristic solutions to larger

problems.
The LP approaches that have been proposed to date appear to share the com-
mon weakness of unpredictability of effectiveness. That is, they produce an
optimal solution quickly on some problems but not on others. The Patterson-
Huber algorithm mentioned above, for example, operating on an IBM 360-67
computer solved some 20-activity problems optimally in seconds while other,
essentially similar, problems required 10-20 minutes of computer time. If the
characteristics of those problems which are amenable to solution by LF' could
be identified, the future for LF' approaches might appear brighter than it now
does.
In summary, in spite of some encouraging recent development, LP approaches
still have not progressed to the point where they are capable of routinely solving
the type of large, complex problems easily handled by heuristic procedures.
They have been used with heuristic procedures to determine the best sequence
of activities within the OSS at individual time periods. However, they remain
today primarily an interesting research topic for academicians.
Enumeration Techniques. The second category of optimization procedures
consists of techniques that have appeared on the scene fairly recently. These are
based on enumeration of all possible activity sequencing combinations (i.e.,
schedules) and include the so-called "branch and bound" procedures. These
methods were first applied to constrained-resource project scheduling in the
late 19609, when researchers in the USA and Germany independently demon-
strated that such procedures were capable of optimally solving networks of up
to about 100 single-resource activities or 30 multiresource activities, under
assumption of nonconstant resource availabilities and activity splitting."'"312
More recent developments include procedures which appear capable of con-
sistently producing optimal solutions to larger problems much more efficiently.
The term "branch and bound" (B&B) refers to a generic type of optimization
procedure which involves partitioning a problem into subproblems (branching),
and evaluating these subproblems (computing bounds). The procedure can be
conveniently modeled by the nodes and branches of a tree, to enumerate pos-
sible alternatives in arriving at the best solution. This solutions tree is sometimes
referred to as the "B&B tree." The power of the procedure stems from the fact
that the enumeration process is implicit, not explicit, since some incomplete
schedule alternatives generated early in the process can be identified immedi-
ately as not capable of leading to an improved complete schedule (compared to
some initial starting value), and thus are not pursued in the solution develop-
ment process. B&B procedures that have been developed by different researchers
differ in the branching rules used (for deciding which portion of the tree to ex-
plore next) and in the bounding rules used (for identification of nodes least
likely to produce an improved solution).
 Resource Constraints in Project Scheduling 219 1
I
Table 7-2. Computation Results For Optimal Solution of 233 ~roblems.'~

Network Size Number Mean CPU Variance in

(No. Activities) Solved Time (min.l CPU (min.)"

23 60 0.20 0.07
27 60 0.76 1.02
35 57 2.61 19.17
43 3 5.84 60.75
233

A number of different branch and bound procedures for the project schedul-
ing problem have been developed by various researchers. One of the more effec-
tive to date is that of Stinson, et al., described in Reference 17. The details of
this procedure, which is designed for minimizing project duration, are sum-
marized in Appendix 7- 1, and illustrated on a small sample problem. An optimal
solution for the familiar example problem used throughout this chapter (net-
work of Figure 7-4) is also presented there.
To give some indication of the capabilities of the B&B approach, Stinson's
procedure produced the results shown in Table 7-2 for a sample of 240 artifi-
cially-generated test problems. Out of the total of 240 project-type problems
attempted (job-shop scheduling problems were also solved), optimal solutions
were found for 233, or 97%. The program was terminated short of optimality
for the 7 unsolved problems. In each case computer memory demand, not com-
putation time, was the limiting factor, with 500K bytes of memory storage on an
IBM 370-155 allocated. A representative 4-resource problem in the 43-activity
class is shown in Figure 7-16, with the optimal activity start t i e s .

Figure 7-16 Network for representative B&B problem.

220 11 I Advanced Topics

ACTIVITY DURATIONS. RESOURCES REOUIRED. AND OPTIMUM START TIMES

Act No. Dur. Re& Ree Opt. Start Act No. Dur. Re+ Req. Opt Stan

1 0 0 0 0 0 0 22 5 2 6 6 1 48
2 4 3 2 4 2 0 23 1 3 2 4 1 39
3 6 6 2 3 1 0 24 1 2 6 4 3 43
4 6 5 4 3 2 6 25 1 1 1 2 4 44
5 2 1 3 3 1 6 26 5 6 3 6 1 43
6 2 4 1 6 2 12 27 6 6 2 2 6 55
7 3 4 1 4 5 4 28 3 5 1 5 1 40
8 5 1 2 2 2 12 29 3 3 3 2 4 45
9 7 1 5 2 3 17 30 7 4 3 4 3 48
10 6 6 1 5 3 25 31 5 2 5 3 3 61
11 8 6 1 6 3 17 32 2 1 3 5 5 53
12 2 5 1 6 6 14 33 3 2 3 3 2 '62
13 1 3 4 4 3 24 34 6 4 1 4 1 62
14 7 1 1 1 1 16 35 7 3 5 5 4 55
15 2 1 1 4 1 25 36 1 1 4 5 4 67
16 1 1 1 2 3 31 37 3 1 5 1 2 68
17 9 1 3 5 2 27 38 2 5 3 1 3 67
18 5 4 2 3 1 31 39 2 2 4 3 3 65
19 3 5 1 4 3 36 40 1 6 3 2 3 69
20 3 2 2 5 3 40 41 2 2 6 5 4 71
21 4 3 5 6 5 36 42 3 3 3 2 4 71
43 0 0 0 0 0 74

Resource Limits: 10 units each type.

Figure 7-16 Continued

AU of the implicit enumeration approaches developed to date appear to share

the same disadvantage common to LP optimization procedures: unpredictabil-
ity of computation time from problem to problem. For example, as Table 7-2
shows, the variance in CPU time required by the Stinson approach was quite
large, ranging up to as much as 10 times the mean for a given class of problems.
While some efforts were made in the testing of the procedure to isolate possible
problem characteristics contributing to such large variance, no significant con-
clusions were drawn.

NETWORK CHARACTERISTICS: USE IN

RESOURCE-CONSTRAINED SCHEDULING

The experience gained in resource-constrained scheduling with heuristic and

optimization procedures has clearly shown that the effectiveness of both types
of procedures is strongly dependent upon the characteristics of the particular
problems being solved. Yet surprisingly little is known about the exact nature
of this relationship.
In contrast to the considerable effort that has been expended over the past
 Resour- Constraints in Projsct Scheduling 221 ~
20 years on the development of solution techniques, comparatively little effort
has gone into study of the problems themselves, i.e., the size, shape, complexity,
etc., of project network scheduling problems. This is particularly true of prob-
lems encountered in actual practice.
The lack of an accepted methodology for simply describing networks in mean-
ingful terms has hindered the development of a useful classification scheme for
resource-constrained scheduling applications. For example, except for the num-
ber of activities and resource types, there is today no readily accepted way for
describing the differences/similarities between networks, except in very general
descriptive terms, such as "short and fat," "long and thin," etc.
The need for a system or methodology for describing or measuring the differ-
ence between networks, or changes to a given network problem, has been long
recognized. One use of such a scheme would be in estimating the computational
requirements or general effectiveness associated with a particular optimization
procedure. Another use, in the case of heuristics, would be in predicting which
of several heuristics might be most effective on a particular problem, or in esti-
mating the impact on schedule duration of changes in network structure and/or
resource constraints, when scheduling given networks with a selected heuristic.
Various quantitative factors or summary measures have been tested by differ-

Table 7 3 . Examples of Network Summary ~ e a s u r e s . ~

-- - -- -- -- -

I. Measures that characterize ne w o r k size, shape, and logic

Examples: Length: max. no. of consecutive nodes from beginning to end
Width: max. no of nodes in parallel
No. of arcs
Complexity:
No. of nodes

II. Measures that indicate time characteristicsof the network

Examples: Sum of Activity Durations
Average Activity Duration
Variance in Activity Durations
Critical Path Duration
Total Network Slack
Sum of activity durations
Density:
Sum of durations +total free slack

III. Measures that characterize resource demands/availabilities

Examples: Total Work Content (equivalent to cumulative.resource requirements)
Average Resource Requirement Per Activity
Average Resource Requirement Per Period
Product Moment (a measure of location of predominant influence of the
resource type, in terms of first or second half of original project duration1
Resource Utilization Factor (a measure of resource "tightness," in terms of
requirement versus availabilitiesl
ent researchers in attempts to explain the relationship between problem char-
acteristics and effectiveness of a particular solution technique. In general these
attempts have not been successful. However, pattersonI3 developed a regres-
sion model which satisfactorily predicted the relative effectiveness of selected
heuristic rules for project duration minimization on a sample of both single and
multiproject problems.
A somewhat different scheme of network summary measures is described in
Reference 7. One of the more complete such schemes yet proposed, it is in-
tended to provide both a general scheme of project network descriptive measures
as well as factors for use in resource-constrained scheduling. This approach
divides the domain of possible summary measures into 3 categories. Table 7-3
shows these categories, along with selected examples of summary measures in
each category.
One potential use of such summary measures is for estimating the impact of

Figure 7-17 Project network for management information system.

 Resource Constraints in Project Scheduling 223

ACTIVITY DURATIONS AND UNITS OF EACH RESOURCE TYPE REQUIRED

Activity Type I Type2 Activity Type 1 Type2

No. Duration Resource Resource No. Duration Resource Resource

Figure 7-17 Continued

alternative resource availability levels on project duration, to avoid the necessity

of actual solution by computer. The network shown in Figure 7-17, for ex-
ample, is for an actual project involving the design and implementation of soft-
ware for a computer-based management information system. The resources
required in this case were 2 different types of computer programmers. Project
224 I1 I Advanced Topi-

duration, computed without regard to resource limits, was 75 days. However,

the number of programmers of each type was severely limited. Possible com-
binations of each type of personnel ranged from a low of 3 type 1 and 7 type 2
to a maximum of 5 type 1 and 9 type 2 (activity durations were fuced and did
not change with resource loading).
When the network was scheduled successively under each of these 9 levels of
manpower availability (using a minimum Late Finish T i e heuristic), the project
durations indicated by the solid curve ("scheduled" values) in Figure 7-18 was
obtained.
A regression equation was developed for estimating these scheduled values,
using selected combinations of the types of summary measures illustrated in
Table 7-3. This regression equation was used as the basis for a forecasting model
in the form of an interactive computer program. The program, when given the
standard network data (i.e., durations and logic), resource requirements, and
possible combinations of resource limits of each type, automatically calculated
the necessary summary measures and produced the "forecast" durations shown
in Figure 7-18.

Available men per day (Type 1. Type 21

Figure 7-18 Forecast and scheduled durations for network of Figure 7-17,
 Resource Constraints in Project Scheduling 225

SUMMARY

Techniques for scheduling activities to satisfy resource constraints have devel-

oped significantly in both number and capability over the past 20 years. The
past 10 years, in particular, have seen increased activity in this area.
Considerable progress has been made in the past 10 years in the development
of implicit enumeration procedures for producing optimal resource-constrained
solutions. And because of the continued improvements in computer technology,
these procedures appear within the realm of practical application to modest-
sized problems in special cases. However, the extreme unpredictability of com-
putation time associated with these procedures will most likely prevent their
widespread use until either improved algorithms are developed or ways are found
to exploit problem characteristics by applying them only in instances where they
will be effective.
Heuristic scheduling procedures have also experienced significant progress
over the past 10 years. Today there are literally dozens of large-scale, computer-
based project scheduling packages available, which offer varied and imaginative
heuristics that will handle the largest networks under an extreme variety of
special conditions. Such heuristic procedures are the mainstay of commercial
applications, and will likely continue to be for the immediate future.

REFERENCES

1. Agin, Norman, "Optimum Seeking with Branch and Bound," Management

Science, December 1966.
2. Burgess, A. R. and .I. B. Killebrew, "Variation in Activity Level on a Cyclic
Arrow Diagram," Journal of Industrial Engineering, March-April 1962.
3. Davis, Edward W. and George E. Heidorn, "An Algorithm for Optimal
Project Scheduling Under Multiple Resource Constraints," Management
Science, August 197 1.
4. Davis, E. W and J. H. Patterson, "A Comparison of Heuristic and Optimum
Solutions in Resource-Constrained Project Scheduling," Management Sci-
ence, April 1975.
5. Davis, E. W. and J. H. Patterson, "Resource-Based Project Scheduling:
Which Rules Perform Best?" Project Management Quarterly, Sept. 1976.
6. Davis, E. W., "Project Scheduling Under Resource Constraints: Historical
Review and Categorization of Procedures," AIIE Transactions, Dec. 1973.
7. Davis, E. W., "Project Network Summary Measures and Constrained Re-
source Scheduling," AIIE 'Dansactions, June 1975.
8 . Elmaghraby, S. E., Activity Networks: Project Planningand Control by Net-
work Models, John Wiley & Sons, New York, 1977.
9. Fendley, Larry G., "Toward the Development of a Complete Multi-Project
226 II I Advancad Topics

Scheduling System," Journal o f Industrial Engineering, October 1968;

unpublished PhD thesis, Arizona State University, 1966, same title.
10. Gordon J. H., "Heuristic Methods In Resource Allocation," Proceedings o f
the 4th Internet Conference, Paris, 1974.
11. Johnson, T. J. R., "An Algorithm for the Resource Constrained Project
Scheduling Problem," unpublished PhD Dissertation, Massachusetts Insti-
tute of Technology (1967).
12. Mueller-Mehrbach, H., "Ein Verfahren zur Planung des Optimalen Betriebs-
mitteleinsatzes bei der Terminierung von Grossprojekten," Zeitschrift fuer
Wirtschaftliche Fertingueng, Heft 2 and 3, Feb./March 1967.
13. Patterson, James, "Project Scheduling: The Effects of Problem Structure
on Heuristics Performance," Naval Research Logistics Quarterly, 1976,
Vol. 23, pp. 95.
14. Patterson, James H. and Walter D. Huber, "A Horizon-Varying, Zero-One
Approach to Project Scheduling," Management Science, February 1974.
15. Schrage, L., "Solving Resource-Constrained Network Problems by Implicit
Enumeration-Non-Preemptive Case," Operations Research, 10 (1970).
16. Stinson, Joel P., "A Branch and Bound Algorithm for a General Class of
Multiple Resource-Constrained Scheduling Problems," unpublished PhD
Dissertation, University of North Carolina (1976).
17. Stinson, Joel P., E. W. Davis, and B. Khumawala, "Multiple Resource-
Constrained Scheduling Using Branch and Bound," AIIE Transactions,
September 1978.
18. Talbot, F. B. and J. H. Patterson, "An Efficient Integer Programming Algo-
rithm With Network Cuts for Solving Resource-Constrained Scheduling
Problems," Management Science, July 1978.
19. Talbot, F. B., "Project Scheduling with Resource-Duration Interactions:
The Nonpreemptive Case," Working paper No. 200, Graduate School of
Business Administration, University of Michigan, January 1980.
20. Weist, Jerome D., "Heuristic Model for Scheduling Large Projects with Lim-
ited Resources," Managemen t Science, February 1967.
21. Weist, Jerome D., "Heuristic Programs for Decision Making," Harvard Busi-
ness Review, September-October 1965.
22. Weist, Jerome D. and Ferdinand K. Levy, A Management Guide t o PERT/
CPM, Prentice-Hall, Inc., Englewood Cliffs, NJ, 1969.
23. Woodworth, Bruce M. and C. T. Willie, "A Heuristic Algorithm for Resource
Levelling in Multi-Project, Multi-Resource Scheduling," Decision Sciences,
1975, Vol. 6 , pp. 525-540.

EXERCISES

1. Suppose the resource requirements associated with each activity in the net-
work of Figure 7-4 have changed t o the following levels:
 Reaourca Constraints in Project Scheduling 227

Resource Requirement
Activity Type A Type B
A 3 2
B 2 4
C 3 1
D 4 3
E 4 0
F 2 2
G 3 1
H 4 4
I 6 4
J 4 1
K 5 4
a. Draw the ES and LS resource profiles.
b. Construct the cumulative resource requirements curves.
c. Calculate the lower-bound estimate of the average daily resource require-
ments of each type; use the internal tangent method where appropriate.
d. What is the criticality index for each resource if the following maximum
levels of each are available?
Type A Type B
Case 1 8 8
Case 2 7 7
e. What is the expected minimum project duration with Case 1 resource
limits of d above?
2. Using the Burgess leveling procedure and the revised resource requirements
above for the network of Figure 7-4, level the resource requirements as much
as possible.
3. Using the shortest-job-first heuristic (SIO rule in Table 7-1) instead of the
min-LS rule t o order activities in the OSS, reschedule the network of Figure
7-4 with the basic procedure of Figure 7-13. Use lowest job number in case
of ties.
4. Using the new resource requirements given in (1) above for the network of
Figure 7-4 and resources available of 8 type A, 8 type B, apply the basic
scheduling procedure of Figure 7-1 1.
5. Repeat exercise 4 above, but instead of using the Min-LS rule to order activ-
ities in the OSS, use the "Greatest Resource Demand" (GRD) rule shown in
Table 7-1. Use lowest job number in case of ties.
6. Schedule the example of Figure 7-4 using the Wiest procedure outlined in
Figure 7-15. Assume that only resource type A is required, that the activity
resource requirements of type A shown in Figure 7-4 are the "Normal crew
size" and that the maximum and minimum loadings are one unit more or
one unit less, respectively. Assume type A resources available equal 8 units.
Ignore schedule costs.
7. Verify the project schedule shown in Figure 7-9 using the Burgess algorithm.
APPENDIX 7-1
APPLICATION OF BRANCH AND
BOUND METHODS TO RESOURCE
CONSTRAINED PROJECT SCHEDULING

The description of the branch and bound method here as applied to resource-
constrained project scheduling is based on Stinson, Davis, and ~humawala."*
The version of the problem addressed is minimization of project duration, i.e.,
establishing a schedule of feasible activity start times such that the entire project
is completed in a minimum span of time. It is assumed that (1) resource require-
ments of each activity are at a constant level for each resource type during the
entire interval in which the activity is in progress, (2) the level of availability of
each resource type is constant over the entire span of project duration, and
(3) activities once started are not interrupted.
The B&B procedure for the resource-constrained scheduling problem consists
of creating nodes in the decision tree, which represent unique "partial sched-
ules." Each partial schedule represents scheduling decisions for some subset of
the total number of activities. The partial schedules created are always feasible
(i.e., they satisfy both precedence and resource constraints) and nonredundant
(no two schedules are exactly alike). Furthermore the solutions tree is exhaus-
tive in that every possible feasible project schedule will exist at the terminal
nodes of the tree.
The tree generation process starts with creation of an initial node representing
the set of activities which can be started at the beginning of the project (time 0).
From this point a family of partial schedules is created by branches with new

*Reference numbers are for the references to Chapter 7.

228
 Resource Constraints in Project Scheduling 229

nodes to the tree; each node in the family created from a particular node has in
common with the others all scheduling decisions made previously in creating the
common parent node. However each is unique from the others in that it in-
cludes one new decision involving the scheduling of one or more activities pre-
viously unscheduled. Each branching operation therefore creates only as many
new partial schedules as there are feasible combinations ("feasible subsets") of
activities that may enter the schedule at some point in time, t,. Thus a partial
schedule, PS,, can be visualized as a real project in progress at time t,: Some
activities (the "complete set," C,) will have been completed at t,, others (the
"active set," A , ) may be actively in progress, to be finished at a later date, and
still others may be ready to be scheduled when all predecessor activities have
been completed.
The complete B&B procedure involves simultaneous tree generation and
"pruning," to prevent growth beyond unmanageable bounds. However the pro-
cedure is best understood by first showing the complete (i.e., unpruned) tree
for a sample problem; then techniques for pruning the tree will be described.
The small 4-activity project of Figure 7-19 will be used as an example. The
B&B solutions tree generated for this project is shown in Figure 7-20. The first
node (partial schedule) represents the only feasible scheduling decision at time
zero, which is to place activity 1, the dummy start activity, in an "active" status.
When activity 1 is completed (also at time zero) there are three possible schedul-
ing decisions which are feasible (i.e., 3 possible feasible subsets): Schedule activ-
ity 2 by itself (node 2), schedule activity 3 by itself (node 3), or schedule both
activities 2 and 3 together (node 4). The latter is allowable only because the
resource requirements of both activities together do not exceed the amount
available.
To illustrate continuation of the process, consider branching from node 2,
where activity 2 is scheduled to start at time zero. The next set of scheduling
decisions will occur when activity 2 is completed (time 1). Thus all scheduling

k Duration

Figure 7-19 Simple network for branch and bound illustration.17

230 11 I A d v a n d Topia

2:o 3; 0 4: 0
121;12.3) 131;IZ. 3) 12.31; (2. 31

I
I 5;l 7;4 8; 1
13); 131 (21; 1-21 (3); 0

1
1
9; 5 10; 5 11;5 12;5 13; 5 14; 4
141; 14) i31;I3.51 15?;13.51 (3.5); {3.51 14); 14) {4): (41

I I I I I
15;9 16; 9 17;9 18;9 19;8
(51; (51 151; 151 [31: (31 (51: 15) (51; (51

16); 161 161; W1

n : Node number F; : Feasible subset S of
t, : Schedule time activities starting at
A, : Active set time t,

Figure 7-20 Complete declsion tree f o r illustrative network."

decisions in the new family of nodes emanating from node 2 will occur at time
1. The feasible alternatives are: Schedule activity 3 alone (node 5) or schedule
activity 4 alone (node 6). The alternative of scheduling both activities 3 and 4
together is not feasible because of resource constraints.
Continuing from node 6 (activity 4 selected to start at time l), there are three
possible alternatives: Schedule activity 3 alone (node lo), schedule activity 5
alone (node 1I), or schedule activities 3 and 5 together (node 12). All three new
nodes have the same "decision time" of day 5, since activity 4 is complete at
time 5.
If node 12 is selected to continue from, it can be seen that both activities 3
and 5 will be completed simultaneously at the end of day 9. At this point the
only possible new decision is to schedule the final (dummy) activity 6. Thus
node 23, a terminal node, is a complete schedule since all project activities have
been scheduled. Node 23 is, in fact, the optimal schedule; its completion time
is 9 days. The Gantt chart for node 23 is shown in Figure 7-21 along with all
nonoptimal complete schedules (terminal nodes 20,21,22, 24, and 25).
The tree of solutions created for this sample problem can be seen to be ex-
 Resource Constraints in Project Scheduling 231

Optimal schedule IPS,,)

Res.
used

Time
la)
Non-optimal schedules
psm
2
1 Activity 3 can
be left-shifted
0
0 1 2 3 4 5 6 7 8 910111213

2
1 Activity 5 can
be leftshifted
0

ps,
2
1 Activity 3 can
0 be left-shifted

2
1 Activity 2 can
be left-shifted
0
0 1 2 3 4 5 6 7 8 9 10111213

If)
Figure 7-21 Complete schedules for illustrative network.17

haustive, and feasible with respect to both precedence and resource constraints.
It is also nonredundant since no two complete schedules are exactly alike. It
offers a framework for evaluating all possible feasible schedules. However, fol-
lowing this process is quite impractical for even modest-sized problems, because
the number of complete schedules rapidly becomes extremely large as the num-
ber of activities increases. The framework offered is thus useful only to the
extent that portions of the solutions tree can be pruned away during the tree
232 11 I Advanced Topics

creation process, in order to keep the tree within manageable bounds. Such
pruning is allowable, however, only if it can be proved conclusively that further
branching from the pruned-away portions of the tree could not possibly lead to
a complete schedule which would be better than some other complete schedule
that already exists or that can be developed from the nonpruned portions of
the tree.

PRUNING THE TREE

Pruning the solutions tree can be accomplished with two basically different
categories of procedures: (I) schedule dominance and ( 2 ) lower bounds. Sched-
ule dominance is relatively more powerful, but operationally more difficult to
implement. We will describe dominance procedures first.

Schedule Dominance

There are several different approaches to determining schedule dominance; all

of them involve some form of schedule comparison. The approach described
here was developed by Stinson16 and utilizes some proofs of Schrage" and
Johnson."
To grasp the essential notions of schedule dominance, look again at the
branchmg process from node 1 of the example tree in Figure 7-20. In branching
from node 1 to create nodes 2, 3, and 4, all possible combinations of activities
2 and 3 were scheduled. Note that node 3 represented scheduling activity 3
alone. Since activity 3 has a duration of four days, all scheduling decisions
"downstream" of node 3 must occur at day 4 or later. Thisincludes activity 2,
which technologically can be scheduled to start simultaneously with activity 3,
is resource-feasible with activity 3, but is only 1 day in duration. Thus, irrespec-
tive of when activity 2 later appears in any schedule emanating from node 3, it
can always be left-shifted to start at time 0. The Gantt chart for node 24
(shown in Figure 7-21), the only complete schedule downstream from node 3,
shows activity 2 starting at day 4. If activity 2 were left-shifted, the schedule
could be shortened by one day and hence replicate the complete schedule of
node 25. Consequently it can be said that the partial schedule of node 3 is dom-
inated by another partial schedule (node 4) and node 3 could have been elim-
inated from consideration in expanding the tree (thereby eliminatingnodes 7,13,
18, and 24). Nodes 10 and 11 could also have been pruned in similar fashion.
Both nodes 2 and 3 in the sample tree involve scheduling only one activity
while both simultaneously are feasible. Yet node 3 is dominated while node 2
is not (clearly, since the optimal path emanates from it).
While node 2 obviously cannot be dominated directly, any immediate suc-
cessor which incorporates activity 3 into the schedule may be dominated. Here
node 5 is dominated because activity 3 can be left-shifted one day in the com-
 Resource Constraints in Project Scheduling 233

plete schedule shown downstream as node 21, making it also equivalent to the
complete schedule of node 25. Thus node 5 is also dominated by node 4, the
parent of node 25.
At this point in the pruning process the surviving portions of the tree are
shown in Figure 7-22. Only one path other than the optimal path remains. This
path cannot be eliminated by the type of schedule dominance illustrated above,
but may be pruned by use of a lower bound.

Lower Bound Pruning

Note that at any point in the project scheduling (i.e., for any partial schedule) a
minimum length schedule for the remaining (unscheduled) activities can be cal-
culated ignoring possible resource conflicts. In other words, there is an inescap-
able duration of the remaining work implied soley by the duration and prece-
dence requirements of the remaining activities. The completion time of this
path constitutes a lower bound on completion time of any partial schedule
emanating from this partial schedule. The authors of Reference 17 call this a
precedence-based lower bound (LBP). If it is not less than the completion time
for a known complete schedule (called the current upper bound), the partial
schedule may be pruned.
Another type of lower bound can be calculated by ignoring precedence con-

Figure 7-22 Reduced solutions tree for illustrative problem.

straints and looking at resource requirements. Any project has a f ~ e work
d con-
tent of each resource type, equal to the sum over all activities of the daily re-
source requirement times activity duration. With a known limit on resource
availability in each time period, a resource-based lower bound (LBR) is clearly
implied. For example, if the project in question requires 100 man-days of work
of some resource type and 7 men of that type are available each day, then before
the project starts we know that its duration must equal or exceed 15 days (y =
14.3 = 15). Similarly, suppose some partial schedule with a decision time of 5
days is being considered, with a total of 75 man-days of unscheduled work re-
maining. If there is an existing possible solution of 15 days and the resource
limit is 7 men per day, then in the 10 days between and including day 5 and day
14 (i.e., one day less than the existing schedule) there are only 70 min-days
available. Thus this partial schedule could not possibly be part of a shorter com-
plete solution and could be eliminated.
In practice both the precedence-based LB and the resource-based LB can be
computationally implemented much more efficiently than the above descrip-
tions suggest. For example, a more convenient way of using critical path infor-
mation is to calculate a late start time (LST) for each activity, based on the best
complete schedule to date. Any partial schedule which leaves an activity in the
set of remaining unscheduled activities on a date greater than or equal to its LST
cannot lead t o an improved solution.
The two LB's mentioned above have been found through computational ex-
perience t o be relatively weak because they consider either resource constraints
and ignore precedence constraints, or vice versa. Use of these two LB's will not,
for example, permit elimination of node 4 in our sample problem.
A stronger LB can be developed which involves simultaneous consideration of
both resource and precedence constraints. This bound, termed a critical se-
quence lower bound (LBC) is described in References 16 and 17. In practice,
all three lower bounds (i.e., LBC, LBR, and LBP) are calculated and the largest
of the three taken as the lower bound for the partial schedule.
Figure 7-23 shows a reduced solutions tree for the familiar 2-resource problem
of figure 7-4. The initial lower bound is 19 (LBR based on resource 1 : =
18.5 = 19); however, a lower bound of 22, based on the heuristic solution of
Figure 7-12, can also be used. Cumulative resource idleness (used in branching,
as described below) is indicated for each partial schedule. The partial schedule
alternatives eliminated by dominance or lower bound pruning are indicated. As
can be seen, the optimum schedule duration is 20 days. Figure 7-24 shows the
resulting bar-chart schedule.

TREE DEVELOPMENT AND BRANCHING RULES

The process of solving the resouce-constrained scheduling problem with B&B

involves the primary sequential steps of (1) branching, or building up the tree,
 1
DOM I B 4 8129.3018 1
IL <C C127. 5119
D(13.2315
& I kM2 j 8C114,
: 2518
DOM t. BD BD(9, 1515

C(21.39111
E(15, 2418
F118,21)8
G(15, 7)9
CE(6. 2118 CE t OOM
CF19. 1818 CF + I L

CEF
8 1 3.6 13,18
C11
G(3, 33112
F17.33112 -F t IL
G
EF(3.33112 -EF t DOM
11 13,613.33
G 12
113. 37)

Schedule Available Cumulative

idleness of
time resources
I /each resource
type
Active jobs and
completion time Active
set

Completion
number time

\
Cumulative idleness
of each resource type
Feasible subset pruned by:
IL = Idleness limit
CP = Critical path bound
DOM =Dominance

Figure 7-23 Reduced solutions tree for network of Figure 7-4 (see legend, next page).
236 11 I Advanced Topics

I
F I
6 6 6 7 7 2 2 2 2 2 2 3 4 4 4 3 3 4 4 4
5 5 5 6 6 6 6 6 6 6 6 6 5 5 5 6 6 5 5 5
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 5 10 15 20
Figure 7-24 20day optimum schedule for sample network.

(2) bounding, or evaluating the nodes, and (3) pruning nonoptimal portions of
the tree. As can be seen from the example above, the more effective the pruning,
the more efficient the procedure becomes. However pruning is also enhanced by
the manner in which the tree is allowed to grow, i.e., the manner in which the
next unpruned node from which t o branch is selected.
The node selection rule used for branching can cause the solutions tree t o
develop in two strongly contrasting ways. One such way is termed "backtrack-
ing." In this scheme the tree develops in depth very rapidly in a restricted area
until a complete new solution is obtained; lateral areas of the tree are not ex-
plored until this restricted area has been fully explored and pruned.
The second way in which the tree can be developed is termed "skiptracking."
In this case the tree tends to grow in breadth more rapidly and downward move-
ment is in more or less uniform fashion laterally.
Skiptracking schemes require significantly more computer storage than back-
tracking procedures, but generally can be solved in less computer time. More
importantly for the constrained-resource scheduling problem, tree pruning by
schedule dominance is possible with skiptracking but not backtracking.
Two node selection rules which result in skiptracking are: Select the node
having the least lower bound, or select the node which has the least total accu-
mulated resource idleness up to the current partial schedule time. Use of any
one node rule alone, however, is not as effective on vely large problems as sev-
eral rules together; this is because many nodes could have the samelower bounds.
A series of partial schedule "attributes," including the three lower bounds de-
scribed earlier, can be grouped into a "decision vector," i.e., a series of tie-
breaking rules, for selecting the next node from which to branch. If no ties
exist for the first attribute in the vector, the other attributes are ignored.
B&B procedures represent perhaps the most promising avenue of mathe-
matically rigorous alternatives for attacking the constrained-resource scheduling
problem. As computer memories have expanded along with significant decreases
in the cost of computer time, these procedures have become less unreasonable to
consider, at least for some small-medium sized problems.
TIME-COST
TRADE-OFF
PROCEDURES

The results of the planning and scheduling stages of the critical path method pro-
vide a network plan for the activities making up the project and a set of earliest
and latest start and fmish times for each activity. In particular, the earliest oc-
currence time for the network terminal event is the estimated "normal" project
duration time, based on "normal" activity time estimates. This state of the over-
all project planning and control procedure is depicted in box (3) of Figure 1-4
(Chapter 1). The purpose of this chapter is to consider the question raised in
the next step, i.e., whether the current plan satisfies time constraints placed on
the project, or in general, to consider the relationship between project duration
and total project costs.
Time constraints arise in a number of ways. First, the customer might con-
tractually require a scheduled completion time for the project. Then, the origi-
nal time constraint might change after a project has started, requiring new proj-
ect planning. These changes arise because of changes in the customer's plans;
or, when delays occur in the early stages of a project, the new expected comple-
tion time of the project may be too late. The most interesting time constraint
application, and the one which was the basis for the development of the "CPM"
time-cost trade-off procedure by Kelley and ~ a l k e r , ' ~ 'arises
~ ' when we ask for
the project schedule that minimizes total project costs, direct plus indirect. This
is equivalent to the schedule that just balances the (indirect) marginal value of
time saved (in completing the project one time unit earlier) against the (direct)
marginal cost of saving it. This situation occurs frequently, for example, in the
238 11 I Advanced Topics

major overhaul of large systems, such as chemical plants, paper machines, air-
craft, etc. Here the value of time saved is very hgh,and furthermore it is known
quite accurately. In this application, the crux of the problem amounts to devel-
oping a procedure to find the minimum (marginal) cost of saving time. This as-
sumes, of course, that some jobs can be done more quickly if more resources are
allocated to them. The resources may be men, machinery, and/or materials. We
will assume that these resources can be measured and estimated, reduced to
monetary units, and summarized as a direct cost per unit time.
Thus, the main purpose of this chapter can be stated as the development of a
procedure to determine activity schedules to reduce the project duration time
with a minimum increase in the project direct costs, by buying time along the
critical path(s) where it can be obtained at least cost.
This procedure can be applied informally in a very simple manner. For exam-
ple, consider the network presented by ~ a v i s ?and shown in Figure 8-1. The
critical path is composed of activities 0-1, 1-2, 2-4, and 4-5, with normal dura-
tion times of 4, 6, 5, and 7, respectively, giving a project duration of 22 days.
The other paths through the network have positive float values indicated along
the arrows. Also, the third value inside the parentheses indicates the cost to
reduce the duration of the activity by one day. For example, activity 0-1 has
a normal duration of 4 days, and it can be reduced to 3 days at a cost of $70.
It is easy to see in this example that the cheapest way to compress this project
is to add "resources" to activity 1-2 and reduce its duration by 1 or 2 days.
The net result in this case would be a project duration of 21 days at an extra
cost of $50, or a 20 day duration at an extra cost of $100. The effect of a 2
day reduction in activity 1-2 is noted directly on the network by the changes
made in the forward pass computations only. At this point it is interesting to
note that there are now three critical paths; 0-2-4-5 and 0-1-4-5 have been
added to the original critical path 0-1-2-4-5. Further reductionsin the project
duration are still possible, but now one must determine the economical way to
cut all three paths simultaneously.
Systematic methods of carrying out the above procedure will be taken up
in this chapter. First, the original CPM approach of Kelley and ~ a l k e r , ' ~ ~ "
based on simple linear time-cost tradeoff curves for each activity, will be pre-
sented. Then, a heuristic hand computational procedure will be presented,
based on these same assumptions. Finally, an extensive treatment of the use of
mathematical programming to give exact solutions to this problem is presented
in Appendix 8-1.
It is assumed in all of the procedures to be developed in this chapter that un-
limited resources are available. If this is not the case, or if personnel must be
paid up to some fixed resource requirement level, whether they are needed or
not on a particular day, then the procedures described in Chapter 7 on resource
allocation may be more appropriate than those described here. Although this
 Time-Cost Trade-Off Promurea 239

4'

(8-6-801

Key:
-
Normal
time \(:I?, J
Crash Cast

y& EF

'Changes in the forward pass computations reflect a compression of activity 1-2 from 6 to 4 days.

Normal "Crash"
Time slope

3 days
6 days
4 days
7 days 600
1 day 1.100 200

150
4 days
3 days
240
150
.
6M) 6 days 750 150
$4.280

**This activity cannot be expedited.

Figure 8-1 Example of elemantaw timecost trade-off procedure showing compression of

project from 22 to 20 days.

assumption of unlimited resources is not often completely satisfied, there are

situations where it is satisified to the extent required here. For example, the
project in question may be a high priority project that will draw personnel from
a large number of low priority or deferrable work activities, so that there are
effectively unlimited resources.

THE CRITICAL PATH METHOD (CPMI

OF TIME-COST TRADE-OFFS

The development of the basic CPM time-cost trade-off procedure is based on a

number of special terms which are defined below and are further shown in
Figure 8-2.
240 11 1 Advanced Topics

Activity direct costs

I "Crash" time-cost point. coordinates (d,C,l

Actual cost curve is assumed to be "convex".

"Crash"
cost C,
- I . e., a straight line connecting any two points
on the curve lies entirely above the curve.

Linear approximation

"Normal"
. . "Normal" time-cost
point, coordinates

i.
- I -- - - - - - - - - - -

[ i I
Activity

td
1 duration
time
"Crash" time D
"Normal" time

Figure 8-2 Activity timecost t r a d e a f f input for the CPM procedure.

Definition:

Activity direct costs include the costs of the material, equipment, and direct
labor required to perform the activity in question. If the activity is being per-
formed in its entirety by a subcontractor, then the activity direct cost is equal
to the price of the subcontract, plus any fee that may be added.

Definition:

Project indirect costs may include, in addition t o supervision and other cus-
tomary overhead costs, the interest charges on the cumulative project investment,
penalty costs for completing the project after a specified date, and bonuses for
early project completion.

Definition:

Normal activity timecost point. The normal activity cost is equal t o the mini-
mum of direct costs required to perform the activity, and the corresponding ac-
tivity duration is called the normal time. (It is this normal time that is used in
the basic critical path planning and scheduling, and the normal cost is the one
usually supplied if the activity is being subcontracted.) The normal time is actu-
 TirnaCogt Trade-Off Procedures 241

ally the shortest time required to perform the activity under the minimum direct
cost constraint, which rules out the use of overtime labor or special time saving
(but more costly) materials or equipment.

Definition:

Crash activity timexost point. The crash time is the fully expedited or mini-
mum activity duration t i e that is technically possible, and the crash cost is
assumed to be the minimum direct cost required to achieve the crash perfor-
mance time.
The normal and crash time-cost points are denoted by the coordinates
(D, CD) and (d, Cd),respectively, in Figure 8-2. For the present, it will be as-
sumed that the resources are infinitely divisible, so that all times between d and
D are feasible, and the time-cost relationship is given by the solid line. I t will
also be assumed that this curve is convex, (defined below), and can be adequately
approximated by the dashed straight line. These assumptions are relaxed in the
treatment of this problem given in Appendix 8-1.

ACTIVITY TIME-COST TRADE-OFF INPUTS

FOR THE CPM PROCEDURE

The basic activity inputs to the CPM procedure have been illustrated in Figure
8-2, where it is assumed that the time-cost trade-off points lie on a continuous
linear or piece-wise linear decreasing curve. A piece-wise linear curve is illus-
trated in Figure 8 3 . I t is further assumed that the activities are independent,
in the sense that buying time on one activity does not affect in any way the
availability, cost, or need to buy time on some other activity. This assumption
would, for example, be violated if a special resource could be obtained to speed
up simultaneously two separate activities in the network, since this would mean
that buying time on one activity would automatically include the other.
If the actual tie-cost relationship departs significantly from the assumed
straight line, but is "convex," then it may be necessary to fit the actual cost
curve with a series of straight lines, as shown in Figure 8-3. A convex curve is
one for which a straight line connecting any two points on the curve lies entirely
above the curve; similarly, if the straight line lies entirely below the curve, it is
said to be "concave." These definitions are illustrated in Figure 8-4; the need
for the convexity assumption will be described below.
In Figure 8-3, the actual cost curve, which is convex, is approximated by a
piece-wise linear curve, each piece being treated as a separate activity or pseudo-
activity. In the project network, the actual activity, A, isreplaced by thepseudo-
activities, A , , A , , and A , drawn in series as shown in Figure 8-3. In this illus-
tration, the approximation is with three pieces; however, the procedure can be
242 11 / Advanced T o p i c s

Activity direct costs nCrmhw time-cost

Network representation

--Actual activity

..Pseudo activities

I c o n slopel = ("crash" cost - "normal" cost) I("normal" time - "crash" time)

Preudo-activities

D = d + A t , +At,+At, C, = $ (C, + AC, + AC, + AC,I

Figure 8-3 Piece-wise linear a p p r o x i m a t i o n t o convex t i m e c o s t curves using pseudo-

activities.

extended in an obvious way to any number of pieces. The coordinates of the

normal and crash time-cost points for each pseudeactivity are given in the table
at the bottom of the figure, where it can be noted that the sum of the three
pseudo-activities, A , , A 2 , and A,, gives the whole activity, A, and the sum of
the coordinates of the normal and crash points for the pseudo-activities gives

Convex Concave Neither concare

Figure 8-4 Illustration of convexity and concavity.

 TimeCoat Trade-Off Procedunr 243

Figure 8 6 Example of a concave activity timecon trade-off curve.

the coordinates of the same points for the whole activity, i.e., (D, CD) and
(d, Cd). The reason for the convexity requirement can be explained heuristically
in terms of Figure 8-3. If the activity is currently scheduled at its normal time,
D, then physically, pseudo-activity A, must be augmented first, then A2, and
finally A , . Since the CPM computational procedure effectively searches the crit-
ical activities to find the one that can be augmented the cheapest, it will naturally
choose the pseudo-activities in the proper order, i.e., A,, A 2 , and finally A , ,
since the cost slopes increase as one goes from A, to A, to A , , for any convex
curve. However, if the time-cost curve was not convex, then the cost slopes may
be lowest for A , , and highest for A , , as shown in Figure 8-5. In this case, the
CPM computational procedure would augment the activities in a sequence that
would not be physically meaningful, i.e., in the order A , , A 2 , and finally A3.

The CPM Computational Procedure

The CPM computational procedure chooses the duration times for each activity
so as to minimize the total project d i i c t costs and at the same time satisfy the
constraints on the total project completion time and on the individual activi-
ties, the latter being dictated by both the logic of the project network and
the performance time intervals (d, D) established for each activity.
For example, again consider the simple project presented in Figure 8-1. Ap-
plying the " CPM procedure, we would start by setting all activity durations at
their "normal" value. This gives a project duration of 22 days as determined
by the critical path consisting of activities 0-1, 1-2, 2-4, and 4-5. The associated
total direct cost of project performance is $3050, as indicated in Figure 8-6.
Note that this cost could be increased t o $3,870 through unihtelligent decision
making by "crashing" all activities not on the critical path, with no decrease in
project duration. Between these upper and lower cost values for a project dura-
tion of 22 days there are several other possible values, depending upon the num-
ber of noncritical activities crashed.
If all activity durations are set at "crash" values, the project duration can be
decreased to 17 days, with a total cost of $4,280, as shown by the extreme up-
per left point of Figure 8-6. This corresponds to the practice of speeding up a
244 11 I Advanced Topics

Line of maximum
direct costs
Crash all

+.
0
0

direct costs
3000 - 3050
I I I I I
17 18 19 20 21 22
Days

Figure 8-6 Project duration vs. direct cost for sample network in Figure 8-1.

project by requiring crash times across the board (everybody works overtime),
instead of along the critical paths only. In place of this, a duration of 17 days
could also be achieved at lower cost by not "crashing" activities unnecessarily.
Thus, activity 0-2 can be set at 7 instead of 6 , activity 1-4 at 8 instead of 7,
and activity 2-3 at 4 instead of 1. With all other activities set at crash values,
the associated cost of performance for 17-day project duration is reduced to
$3,570. This value is the lowest possible value for 17-day project duration,
as will be shown below. It is the optimal " CPM solution for a 17-day schedule.
Between 22 and 17 days there are several possible values of project duration,
as shown in Figure 8-6. For each such duration there is a range of possible cost
values depending upon the durations of individual activities and whether activi-
ties are crashed unnecessarily or not. Figure 8-6 shows the curve of both maxi-
mum and minimum costs and the region of possible costs for each duration be-
tween these curves.
In this simple example the minimum direct cost curve is easily determined by
trial and error, or by the hand application of a heuristic procedure such as that
given below. But in more realistic cases consisting of several dozen or more ac-
tivities, such trial and error determination becomes extremely tedious, if not
impossible. In these situations, systematic computation schemes, usually rnathe-
matical programming, must be used to determine the value of the minimum cost
curve for every possible level of project duration. The details of these proce-
dures are discussed in the appendix to this chapter, while their application is
illustrated in Figure 8-7.
Suppose the project duration is under management control, such as in the
case in a plant overhaul. Assuming adequate resources are available, the rational
choice of a project schedule would be made so as to minimize the total project
costs. Then, one would select the project duration schedule indicated in Figure
 TimeCort Trade-Off Procedures 245

Project duration

Figure 8-7 Determining project schedule for minimum total cost.

8-7, corresponding to the minimum value of the total cost curve. The critical
ingredient of this process is the minimum direct cost curve, which was not really
available prior to the inception of this "CPM" optimization procedure.

HAND COMPUTATIONAL PROCEDURE

The introduction of this chapter included a discussion of the hand compression

of the project shown in Figure 8-1. The activity to be shortened was merely
selected by visual inspection of the network. In this case the most economical
alternative was quite obvious. However, as the compression process continues,
the number of critical paths increases, and it very quickly becomes difficult t o
evaluate all of the possible alternatives. For this reason, it is useful to adopt
some procedure which can reasonably be followed by hand. While these heuris-
tic hand procedures do not always lead to optimal solutions, they usually come
quite close.
The 8 step hand procedure presented below is a slight modification of the
method developed by Siemens." The key element of this procedure is the cost
slope and time available, illustrated in Figure 8-3. The cost slope will be denoted
by Cij for an arbitrary activity (i - j).

Cost Slope = Cij = (Cd - Co)ij/(D- d)ii

Time Available = TAii =(D - d)ii

An "effective" cost slope, ECij, is then defined by Siemens as the cost slope
divided by the number of inadequately shortened paths, Nij, which contain
activity (i - 1).
246 11 / Advanced Topics

~ i ~ u r e 8 - 8Application of modified Seimens algorithm t o the network problem shown

in Figure 8-1, for thecompression of the project duration from 22 t o 1 7 days.

The procedure described below chooses from among all available activities to
be shortened, the one with the lowest effective cost slope. This heuristic will
usually come quite close to the optimal solution. Each step of this procedure
is illustrated in Figure 8-8.

Modified Siemens Algorithm

1. Prepare the project network and time estimates, and list in columns all paths
through the network whose expected lengths are greater than the desired
(scheduled) project duration, T,. The length of a path is merely the sum of
the durations of all activities on the path in question. Also note at the bot-
tom of each path column (row marked iteration O), the time reduction that is
required, i.e., expected path length minus T,.*

*A systematic procedure for r i d i n g the K shortest paths through a network is given b y

yenI6. A useful alternative procedure would be t o modify a slack sort of the project
activities so that each slack path extends from the initial t o the terminal network event.
 TimeCost Trade-Off Procedures 247

2. List (in rows) all activities present in at least one of the listed paths noting
for each activity its cost slope, Cii, and time reduction available, TAij.*
3. Compute the effective cost slopes, ECii, and record them in the column
headed iteration 1.
4. For the path(s) with the most remaining time reduction required, select the
activity with the lowest effective cost slope. Break ties by considering the
following ordered list:
4.1. Give preference to the activity which lies on the greatest. number of
inadequately shortened paths.
4.2. Give preference to the activity which permits the greatest amount of
shortening.
4.3. Choose an activity at random.
5. Shorten the selected activity (i - j ) as much as possible, which will be equal
to the minimum of the following:
5.1. the unallocated time remaining for the selected activity (i - 13,or
5.2. the smallest demand of those inadequately shortened paths containing
the activity (i - j).
6 . Sell back, or deshorten, as much time as possible on paths that have been
overcut, as long as this action does not cause any new paths to become in-
adequately shortened.
7. Stop if all paths have been adequately shortened. If not, recalculate those
effective cost-slopes where any of the following have occurred:
7.1. a path which was inadequately shortened prior to this iteration, has
been adequately shortened, or
7.2. all unallocated time for the activity just shortened has been consumed
and there are one or more additional cost-slope/supply pairs for this
activity (see the footnote to Step 2).
8. Retum to Step 4.

Example Problem

The sample problem shown in Figure 8-1 will be used to illustrate the above
algorithm. The results of Steps 1, 2, and 3 are shown in Figure 8-8 for the
case where a 17-day project duration is required. There are 5 paths through
the network, but only the 3 listed in Figure 8-8 exceed the desired project
duration of 17 days. Also, activities 2-3 and 3-5 have been omitted because
*1f a piecewise linear approximation is being used, as in Figure 8-3, then the data pairs
( C p TAG) must be given for each linear segment, with increasing cost slopes in columns
golng from left to right. Only one cost-slope/supply pair for each activity can be con-
sidered during any iteration of the following recurring steps. Also, these pairs are con-
sidered sequentially left-to-right as they are used up in the shortening process, and from
right-teleft if deshortening occurs.
248 II 1 Advanced Topics

they are not present in any of the 3 paths which require shortening. The cross-
hatching in Figure 8-8 is used to denote the activities that are not present in
the path indicated at the head of the column.
The repetitive portion of this algorithm then begins by applying step 4,
which leads t o cutting activity 0-1 by 1 day in the first iteration. This activity
was chosen because it has the lowest effective cost slope of the 4 activities on
path 0-1-2-4-5, which is the path currently requiring the greatest amount of
shortening. This change is shown by placing a 1 in the two cells corresponding
to the row for activity 0-1 and the two path columns which contain this ac-
tivity; the "Time Reduction Available" for this activity is updated from 1 to 0;
and, finally, the "Remaining Time Reduction Required" is updated in the row
for iteration 1, to reflect that paths 0-1-2-4-5 and 0-1-4-5 have been reduced
from 5 and 3 to 4 and 2, respectively. A total of 6 iterations are required to
reach a final solution, which requires an augmentation of the direct activity
costs of $520, for a total of $3570. It should also be noted in Figure 8-8, that
17 days is the minimum or "crash" project duration, because all activities on
the first path (0-1-2-4-5) have zero "Time Reduction Available."
A summary of repeated applications of this algorithm to cover the entire
range of possible project compressions is shown in Table 8-1. The results shown
here essentially give the project schedule, activity by activity, for a given project
duration. It can be shown that each of these solutions is "optimal," in the sense
that no other solution will give lower total direct costs. This will not always oc-
cur with this heuristic, based on lowest effective cost slope, but solutions ob-
tained this way will usually be very good. These results were plotted in Figure
8-6 above, where they form the line of minimum direct costs. As stated before,
this is the signal contribution of the " C P M procedure. It permits the specifica-
tion of the optimal (total cost minimization) schedule, as shown in Figure 8-7.

ECONOMIC IMPLICATIONS OF PROJECT TIME-COST TRADE-OFF

The concept of project time-cost trade-off could have considerable economic

importance, considering the extremely large and costly systems being developed
today, over extended periods of time. For example, facilities for electric power
generation, mass transportation, sophisticated weapons systems, oil production,
etc. In each of these cases, the system costs are in the billions of dollars, and
the development times are in excess of 5 years. Alternative project schedules
could have considerably different total costs. An analysis of this problem by
Cukierman and Shiffer3 is given below.
Consider the project duration denoted by T,as a variable to be selected by
management on a rational basis. Factors to be considered include the following,
Table 8-1 Summary of Hand Computation Solutions to the Example in Figure 8-1.

Project Duration (Days)

- -

17 18 19 20 22
mount Amount Amount Amount Amount
ActiviN Shortened Cost Shortened Cost Shortened Cost Shortened Cost Shortened Cost

Total Excess -
Direct Costs

Base
Direct Costs
- 3050 - 3050 - 3050 - 3050 - 3050

Total - 3570 - 3410 - 3260 - 3150 -

Direct Costs
250 11 I Advancad Topics

which denote present value or discounted values based on an appropriate rate

of return, r.
X(T) = minimal discounted costs of a project plan with delivery
date T.
Z(T) = discounted value to the buyer, of the project completed
at time T.
V(T, . . . ) = discounted value of net payments from buyer to seller,
which depend on T, as well as on the detailed payment
schedule; hence the multivariate notation (T, . . . ) includ-
ing T and other variables.
Now, the net benefits to the buyer, contractor, and their total t o society can
be defined, respectively, as follows:

If you assume that the functions Z(T) and X(T) reflect, respectively, the proj-
ects true present value of benefits and costs to society, then society wants
to maximize Bs. This implies that the socially optimal completion time, Ts,
will be determined by the (first-order) condition dZ/dT = dX/dT, independently
of the V function.* This point (Ts) is noted in Figure 8-9, where the slopes of
Z(T) and X(T) are the same. That is, at the point where the marginal benefits
equal the marginal costs.
From the social point of view, the Vfunction is only a device for transferring
income between the buyer and the seller. But its specification is important to
all parties if we are to insure that the contractor has the incentive to also deliver
the project at time Ts, instead of some other time, such as Tc shown in Figure
8-9. A sufficient condition to achieve the desired result is to have aV/aT=
dZ/dT for all T. Cukierman and Shiffer3 point out that this will not be accom-
plished by most project payment plans, V(T,. . . ). For example, periodic prog-
ress payments proportional to contractor expenditures, or agreed upon pay-
ments to be made upon achievement of project milestones, will not achieve the
required incentive. A plan that could achieve the desired result, for example,
would be payments executed at predetermined dates, independently of the
progress of the project, plus an annual fine equal t o the discount rate r, times

~ derivative of Z with respect to T, that is, the rate of change ofZ with respect
* d ~ / isd the
to T.
 TimeCost Trades Procedures 251

Figure 8-9 The socially optimal (Ts)and the contractor's best (Tc) delivery times.

the value to the buyer of the completed project at the time of completion, paid
over the life of the project.
This is an interesting, perhaps academic, consideration of an extremely impor-
tant problem from an economic point of view. Of the three functions in Figure
8-9, X(T) is the most difficult to obtain. The " CPM time-cost trade-off con-
cept could be very helpful in this endeavor to determine a socially optimal proj-
ect schedule. I t should be added that in the determination of X(T), the avail-
ability of resources must also be considered. This could be done, for example,
by trial and error simulation. First, assume a trial set of activity time-cost trade-
off function inputs, and determine the corresponding "CPM" project schedules
for various project durations. The resource level vs. time profdes for these
schedules can then be smoothed by the methods in Chapter 7, and finally, this
smoothed level can be judged as to whether it is feasible or not from a resource
staffing point of view.
It appears that the project timecost trade-off concept has more to offer than
the applications t o date indicate. A real understanding of this concept may lead
to improved solutions to large scale systems problems currently facing society.

SUMMARY

The general philosophy of the time-cost trade-off problem has been presented in
this chapter, along with a hand computation procedure, and the "CPM proce
252 11 I Advanced Topics

dure based on continuous linear time-cost curves and a rigorous computational

algorithm.
An application of the "CPM" procedure in a computerized project manage-
ment system is described by Benson and Sewall.' Their system, called PLANIT,
includes the "CPM" procedure as a subroutine, which they call FASNET. It is
applied in the project planning stage to determine project time-cost alternatives,
and also in the project control stage, as needed to make up lost time. It should
be added that this program also incorporates input constraints on resource avail-
ability as described in Chapter 7. It represents one of the few project manage-
ment systems that incorporate both of these project planning tools.
Embellishments of the basic "CPM" problem are presented in the appendix
to this chapter. First, extensions to the objective function include the work
of E l m a g t ~ r a b ~who
, ~ considered completion schedules, not only on the project
end event but also on an arbitrary set of milestone events throughout the net-
work. He has developed an efficient network flow algorithm to determine the
project/activity schedule which minimizes the sum of the total activity direct
costs plus the penalties for tardy completion of all of the milestone events. An-
other extension of this type was by Moore.13 He used goal programming to con-
sider multiple objectives, such as completion times for milestone events, operation
within a specified budget, etc.
In addition to these embellishments of the objective function, a number of
authors have devised ways of handling more complex activity time-cost trade-
off relationships. These include concave curves in place of the usual convex
assumption. Also, time-cost relationships consisting of a set of feasible time-cost
points, or a combination of feasible points and continuous curves are described
in the appendix t o this chapter.

REFERENCES

1. Bensen, L. A. and R. F. Sewall, "Dynamic Crashing Keeps Projects Moving,"

Computer Decisions, Feb. 1972, pp. 14-18.
2. Charnes, A. and W. W. Cooper, "A Network Interpretation and a Directed
Subdual Algorithm for Critical Path Scheduling," Journal o f Industrial
Engineering, Vol. 13, No. 4 (1962), pp. 213-218.
3. Cukierman, A. and 2. F. Shiffer, "Contracting for Optimal Delivery Time
in Long-Term Projects," The Bell Journal of Economics, Vol. 7, NO. 1,
Spring 1976, pp. 132-149.
4. Davis, E. W., "Networks: Resource Allocation," Industrial Engineering, Vol.
6, No. 4, April 1974, pp. 11 7-1 20.
5. Elmaghraby, S. E. and P. S. Pulat, "Optimal Project Compression with Due-
Dated Events," Naval Research Logistics Quarterly, Vol. 26, No. 2, June
1979, pp. 331-348.
6. Elmaghraby, S. E., Activity Networks: Project Planning and Control b y
Network Models, John Wiley & Sons, New York, 1977.
 TimsCost Trade-Off Procedures 253

7. Fondahl, J. W., "Can Contractors Own Personnel Apply CPM Without Com-
puters," The Constructor, November 1961, pp. 56-60, and December 1961,
pp. 30-35.
8. Fulkerson, D. R., "A Network Flow Computation for Project Cost Curves,"
Management Science, Vol. 7, No. 2, January 1961, pp. 167-179.
9. Hindelang, T. J. and J. F. Muth, "A Dynamic Programming Algorithm for
Decision CPM Networks," Operations Research, Vol. 27, N o . 2, March-
April 1979, pp. 225-241.
10. Kelley, J. E. and M. R. Walker, Critical Path Planning and Scheduling, Pro-
ceedings of the Eastern Joint Computer Conference, December 1959, pp.
160-173.
11. Kelley, J. E., Jr., "Critical Path Planning and Scheduling: Mathematical
Basis," Operations Research, Vol. 9, No. 3 (1961), pp. 296-320.
12. Meyer, W. L. and L. R. Shaffer, Extensions of the Critical Path Method
Through the Application of Integer Programming, Report issued by the
Dept. of Civil Engineering, University of Illinois, Urbana, Ill., July 1963.
13. Moore, L. J., B. W. Taylor 111, E. R. Clayton, and S. M. Lee, "Analysis of
a Multi-Criteria Project Crashing Model," American Institute o f Industrial
Engineering Trans.,Vol. 10, No. 2, June 1978, pp. 163-169.
14. Panagiotakopoulos, D., "A CPM TimeCost Computational Algorithm for
Arbitrary Activity Cost Functions," INFOR, Vol. 15, No. 2, June 1977,
pp. 183-195.
15. Siemens. N.. "A Simple CPM Time-Cost Tradeaff Algorithm," Management
~ c i e n c e , . ~ o17,
l . NO -6, Feb. 1971, pp. B-354-363.
16. Yen, J. Y., "Finding the K Shortest Loopless Paths in a Network," Manage-
ment Science, Vol. 17, No. 11, July 1971, pp. 712-716.

EXERCISES

1. Given below are the network data and the timetost trade-off data for a small
maintenance project.

Table 8-2

Predecessor Normal Crash Cost Slope

Job (Activily) Jobs (daysj (dollarsj (days) (dollarslday)

A none
B none
C none
D A
E C
F A
G 6, D
Total
254 11 / Advanced Topics

Assume the indirect costs, including the cost o f lost production, and asso-
ciated outage costs, supenision, etc., t o be as follows.

Project duration (days) 12 11 10 9 8 7

lndirectcosts(dollars) 900 820 740 700 660 620

Table 8-3

Cost (dollars)
Activity Predecessor Description of Time (hours) Nor-
No. Activity Nos. Activity a m b mal Crash

101 inspect & measure pipe

102 devlp. cal. mils. list
103 make drawings of pipe
104 deactivate line
105 procure calender parts
106 assemble calender work
crew
107 devlp. matl. list (pipe)
108 deactivate calender
109 procure valves
110 procure pipe
111 assemble work crew
(pipe)
tie off warps
prefab pipe sections
erect scaffold
disassemble calender
empty & scour vats
remove old pipe
repair calender
position new pipe
lubricate calender
position new valves
weld new pipe
reassemble calender
adjust & balance calen-
der
insulate pipes
connect pipes to boiler
connect pipes to calen-
der
refill vats
remove scaffold
pressure test
tie in warps
activate calender
clean up
 TimeCost Trade-Off Procedures 255

Assuming any integer times between the normal and crash activity times are
feasible, use the hand computational time-cost trade-off procedure to show
that the total costs are $1510, 1470,1430, 1470, 1530and 1620 for 12, 11,
10,9, 8, and 7 day project durations, respectively.
2. Given in Table 8-3 are the data for a steam calendar and pipeline maintenance
project. Three sets of times are given under columns headed a, m, and b. In
this problem only the first two columns will be used: the times under the col-
umn headed a are the "crash" activity performance times while those under
the column headed m are the "normal" performance times.
a. Draw the network and make the basic scheduling computations using
activities-onarrows or activities~n-nodes.
b. Identify the critical path.
c. Indicate, in step-by-step detail, the activity augmentations required t o re-
duce the project duration to 248 hours while keeping the total project
direct costs at a minimum. Assume that only the normal and crash times
are feasible.
3.*In the appendix to this chapter, the linear programming formulations are
given for activity time-cost trade-off curves that are either continuous, or
a collection of feasible timecost points. These formulations can be extended
to cover the case where combinations of these two types of trade-off func-
tions occur. An analysis of this problem indicates that the transitions can
occur in three distinct ways as shown in Figure 8-10, i.e., any combinations
of continuous curves and discrete points can be handled by appropriate com-
binations of these three transitions. Write out the Linear programming for-
mulation for each of these transitions.

Discontinuous wst curves

Activity duration time

Figure 8-10

4.*Consider the simple project network shown in Figure 8-11, along with the
accompanying activity time-cost trade-off curves. Using the linear (integer)

*This problem pertains to material taken up in the appendix to this chapter.

256 11 / Advanced Topics

programming techniques described in the appendix to this chapter, write the

linear programming formulation of this problem, and determine the schedule
of activity duration times (yli) which will minimize the total direct project
costs for a project duration constraint of h = 10 time units.

-., Activity 1-2 Activity 1-3

Activity duration time

-
v

-
.A

.-
.-
>

.-
4
2
Activity 2-3

3
:a/- - :
Activity

2
I
2-4

I
+':[-;*
Yzs
Activity 3.4

2 5 "34

Activity duration time

Figure 8-1 1
APPENDIX 8-1
APPLICATIONS OF
MATHEMATICAL PROGRAMMING
TO THE DEVELOPMENT OF
PROJECT COST CURVES

The purpose of this appendix is t o present a review of the current applications

of mathematical programming techniques to the problem of generating project
cost curves and the various extensions of this basic problem The treatment
given here will assume that the reader is familiar with the basic linear program-
ming formulation of an optimization problem.
First, the basic network scheduling (forward pass) computations will be
viewed as a problem in linear programming; this approach was published by
Charnes and C ~ o p e r . The
~ time-cost trade-off problem will then be introduced
along with several logical extensions. Finally, this will be followed by the treat-
ment of more complex activity time-cost trade-off functions, along with solu-
tion algorithms.

LINEAR PROGRAMMING FORMULATION OF

BASIC SCHEDULING COMPUTATIONS

The various topics treated in this appendix will be illustrated using the simple
network* shown in Figure 8-12. The numbers appearing along each activity
in this figure give (dii,Dii, Cij). For example, ( d l , , D l , , CI2)= ( l , 3 , 3 ) indi-

*By permission from L. R. Ford, Jr., and D. R. Fulkerson, "Flows in Network," The Rand
Corporation, 1962. Published, 1962, by Princeton University Press.

257
258 11 I Advanmd Topics

Figure 8-12 Illustrative network with data giving crash time. normal time, and cost slope
for each activity.

cates that for activity 1-2, the crash and normal performance times are 1 and 3
time units, respectively, and the slope of the linear time-cost trade-off curve is 3
monetary units per time unit. In the applications, the time unit is chosen so that
the di;s and Dii's are all integer valued, and the monetary unit is chosen so that
the Ci;s are also integer valued.
For purposes of illustrating how the basic forward pass computations can be
formulated as a linear programming problem, assume that all activities in Figure
8-12 are scheduled to be performed at their normal times, i.e., the middle num-
ber of the three numbers given for each activity. Now the project network may
be viewed as a flow network, in which a hypothetical unit of flow leaves the
source, node (event) 1, and enters the sink, node 4. Also, nodes 2 and 3 play
the role of "transhipment" points and, thus, at these nodes there must be a
conservation of flow, i.e., the total flow into node 2 must equal the total flow
away from node 2, and similarly for node 3. The performance time of each
activity, yii, is then interpreted as the time (or cost) of transporting a unit o f
flow from node i to node j. Viewing a project network in this way reduces
the problem of finding the critical path(s) to the determination of the network
path(s) or route(s) from the source, node 1, t o the sink, node 4, which requires
a maximum time (or cost) to traverse. The reader is no doubt well aware of
the fact that there are easier ways of locating the network critical path. The
main purpose of this network flow interpretation of the problem is to show
that it can be formulated as a linear programming problem, a technique which
will prove to be useful later in solving problems which cannot be easily solved
in other ways.
Applying the above network flow interpretation to Figure 8-12 results in
the linear programming formulation given in equation (I), which will be r s
ferred to as the primal problem. In this formulation, yii = 0 or 1 denotes the
absence or presence of a unit flow from node i to node j, a flow which is said to
be along activity i-j.
 Applications of Mathematical Programming 259

PRIMAL PROBLEM

Maximize f [Y]= 3y12 + 4~13 + 5 2 3 + 5 ~ 2 4+ 6 ~ 3 4 (la)

Subject to Y12 + Y I ~ =I (lb)

Constraint equations (lb) and (le) indicate that a unit of flow leaves the source,
node 1, and enters the sink, node 4, while constraint equations (lc) and (Id) re-
quire a conservation of flow at the intermediate nodes 2 and 3. Thus, any set of
yijYswhich satisfy these constraints constitute a path from the source to the
sink, as indicated by the yit's that are equal to one. Then, since the yij's are
equal to one for the activities carrying the hypothetical unit flow, andzeroother-
wise, the objective function, f[Y], gives the sum of the activity duration times
for the chosen path from source to sink. When the objective function is maxi-
mized, the corresponding path, denoted by Y*,is the longest or critical path
through the network; the corresponding value of the objective function will
be denoted by f[Y*]. The solution of this problem is greatly facilitated by
the use of the duality theorem of linear programming as described below.

Formulation of the Dual Problem

From the well known duality theorem of linear programming, it can be shown
that to every linear programming problem, call it the primal, there corresponds
a related linear programming problem which is called the dual of the original
problem. The connection between these two problems is stated in the following
theorem.

Duality Jheorem

Given a linear programming problem, call it the primal, there is always a re-
lated linear programming problem, called the dual, defined as follows, in matrix
notation.
PRIMAL DUAL

Maximize f [XI = [C] [XI Minimize g[W] = [P'] [W]

1 X n nX1 1Xm mX1

Subject to [A] [XI j [PI Subject to [A'] [W] 2 [C]

mXnnX1 mX1 nXmmX1 nX1

[XI 2 [OI [WI 2 [OI

nX1 nX1 rnX1 mX1
If there exists a solution, [X*] , which gives a f d t e maximum value t o f [XI,
nX1
there is always a "coupled" solution [W*] , for which g[W] has a finite mini-
mX1
mum value, equal to f [XI. (A solution to one problem makes the solution t o
the other problem readily available.)
To the above statement of the theorem must be added that while the primal
problem has n variables and rn constraints, the dual problem has the reverse, rn
variables and n constraints. Also, if any of then prim& variables are unrestricted
in sign, then the corresponding dual constraints are equalities, and if the primal
constraints are equalities, then the dual-variables are unconstrained in sign. With
regard to the solutions of the primal and dual problems, if in the solution of the
dual one of its inequality constraints is satisfied as an equality, then the primal
variable corresponding to this dual constraint may be positive, whereas it must
be zero if the dual constraint is satisifed as an inequality.
Applying the Duality Theorem to the above example, one obtains the dual
formulation given in equation (2). Recall that in the primal there were 5 vari-
ables and 4 constraints; hence in the dual problem there are 5 constraints (one
corresponding to each of the primal variables) and 4 variables (one correspond-
ing to each of the primal constraints.) Also, since the primal constraints were
all equalities, the dual variables are all unconstrained in sign.
DUAL PROBLEM

Minimize g[W] = w, - w4 (2a)

Subject tow, - w, 23 (2b)

w2 - w4 2 5 (2e)
w3 - w4 2 6 (2f)
- m < w , <-, - m < w 2 <-, - m < w 3 <-, --<w4 <-
The advantage of the dual formulation t o this problem is now fairly obvious.
Since each dual constraint involves only two variables, they can be solved by in-
spection, if a value is assigned to w, . To see this, consider the constraints writ-
ten in the following equivalent form.
 Applications of Mathematical Programming 261

It will be shown below that w, varies directly with the value assigned tow, ,
and since g [W] is merely the difference between wl, and w4, wl can be as-
signed an arbitrary value without affecting g [W] . Thus, one may let wl = 0,
a choice which makes all of the other wi's, and in particular wq, take on nega-
tive values. Then, to minimize g [W] = w, - w4 =- w4,one must minimize the
absolute value of w4,and at the same time satisfy each of the above constraints.
It is easy to see by inspection that this is achieved by the solution indicated
above, i.e., w: = 0, w: = -3, w: = - 5 and w: = - 11. This will be denoted as
[w*] = (0, - 3, - 5, -1 1). Thus, the optimal value of the dual objective function
is g [w*] = w: - : W = 0 - (- 11) = 11. It is interesting to note the similarity be-
tween the solution of inequalities (2bb) through (2ff), and the conventional for-
ward pass computations. For example the value of w -
: = 5 is the largest abso-
lute value of the two solutions obtained by treating (2cc) and (2dd) as equalities,
or equivalently, the smallest absolute value of w3 which satisfies both of these
constraints. This is analogous to calculating the earliest event time at a merge
point by taking the largest of the earliest finish times of the merging activities,
which are 4 and 5 in this case.
Noting that the second and fourth dual constraints, i.e., those given by equa-
tions (2c) and (2e) are satisfied as inequalities, one knows from the duality theo-
rem that the second and fourth primal variables, i.e., y,, and y, are zero, while
the others may be positive. By checking the primal constraints equations (lb)
through (le), one sees that the optimal solution to the primal problem is given
by [Y*l = ( Y : , , Y T ~ , Y ~ ; , Y ~ ~ , Y=: ~(1,
) 0, 1 , 0 , 11, andthatfry*] = 3 X 1 +
4 X 0 + 2 X 1 t 5 X 0 + 6 X 1 = 11. (Note also that the requirement, f [Y*] =
g[~*= ] 11 is satisfied.) Thus, it is concluded that the network critical path is
comprised of activities 1-2, 2-3, and 3-4, and has a total duration of 11 timeunits,
a value which can easily be verified by the routine forward pass computation.

LINEAR PROGRAMMING FORMULATION OF THE CPM

TIME-COST TRADE-OFF PROBLEM AND EXTENSIONS

The basic formulation of the time-cost trade-off problem will be given first, as-
suming each activity in the network shown in Figure 8-12 has a trade-off curve
of the form shown in Figure 8-13. The advantage of the notation used in this
figure is that it replaces the normal and crash costs by a single cost slope, Cij.
Again letting yij denote the scheduled duration of activity i-j, one can write the
total direct project costs as a function of these variables as follows:

direct = xC(Kii- Cijyii> = K

project costs
- x ZC
i j ilyii

where Xi XI is used to denote the summation over all activities in the project
network. Since the KiiYsare futed constants, whose sum is denoted by K in
262 11 I Advanced Topics

K,, = C,# + C;)D,, ITirnecost slope/= C,,

(Crash1 C.

I i
0
0 d,, (Crash)
-
Dl, (Normal)
v,,

Scheduled activity duration time

Figure 8-13 Timecost trade-off curve nomenclature.

equation (3), the total direct project costs are minimized if one will*

Maximizef [Y] = zCliyii

Subject to Ti + yij - Ti5 0, all ij (4b)

j 5, all ij (4~)
- ~ i j-S dij, all ij (44
T4 - TI 2 h (44
where the Tk's are (unknown) variables denoting the earliest expected time for
node k(k = 1, 2,3, or 4), and h is the (constant) constraint placed on the total
project duration, which is merely T4- TI. Following the CPM convention of
letting T, = 0, the last constraint, equation (4e) would merely become T, 5 A.
The first constraint equation (4b) applies to each activity, i-j,in the network,
of which there are five in this example. These constraints merely state that for
activity i-j, the difference between the earliest node times, Ti and Ti,must be
at least as great as yij, the scheduled duration of activity i-j. Similarly, equation
(4c) applies t o each of the five activities in this example; it constrains the sched-
uled activity duration t i e , yij, to be equal to or less than the normal activity
time, Rii. Finally, equation (4d) constains each yii to be equal t o or greater
than the crash activity time, dq. The variables, Tk, have been omitted from the

*1n this appendix, the symbol Tiwill be used in place of Ei, used elsewhere in this text.
 Applications of Mathematical Programming 263

objective function because their cost coefficients are zero; their role in this for-
mulation is merely to insure that the scheduled values of yij are feasible from
the standpoint of network logic, and to insure that the project duration does
not exceed A.
While this problem could be solved using the simplex method to find the
schedule [Y*]= (yF2, yy3, y:3, y,*,, y;), which satisfies all of the constaints
and at the same time maximizes f[Y], more efficient network flow algorithms
have been developed. These procedures will not be given here. A detailed treat-
ment can be found, for example, in the 2nd edition of this text, or in the origi-
nal reference by Ford and Fulkerson.' An extensive treatment of this subject
can also be found in Chapter 2 of the text by Elmaghraby.6 The development
of this procedure was motivated by considering the dual of the problem given
above by equation (4). The special structure of the nonzero elements in the dual
constraint matrix led to the discovery that this problem could be viewed as a
network flow problem. A network flow algorithm was then devised t o solve the
dual problem, which is much simpler than the conventional simplex method.
Thus, this time-cost trade-off problem, referred to as the " CPM problem, can
be solved very efficiently on a minicomputer, even for large networks. The out-
put of these programs is essentially of the form shown above in Figure 8-8 and
Table 8-1 for the modified Siemens hand computation procedure.
A natural extension of the "CPM" problem, as stated by eq. (4), was devel-
oped by Elmaghraby.' It allows scheduled completion time, Sk,to be placed
on a set of milestone events in the network (k = 1 , 2 , . . . , K), where K is as-
sumed to denote the project terminal event. This problem formulation also in-
cludes penalty rates, pk, that are invoked for tardiness in meeting the schedules,
S k , placed on the project milestone events. This might be applied, for example,
in the development of a mass transit system, a housing project, a shopping cen-
ter, etc., where the facility can be completed and utilized in stages, as defined
by the set of K milestone events. The problem then is to determine the subset
of activities whose durations are to be shortened, and the amount of that shor-
tening, in order to incur the smallest total cost, i.e., cost of shortening plus the
cost of tardiness. The linear programming formulation of this problem can be
written as follows:

Subject to Ti+ y q - Ti $ 0; all ij (5b)

yij IDij; all i j (5~)
264 11 I Advanced Topig

In this formulation, the constraint set (5e) replaces the single constraint (4e)
in the " CPM problem. In these constraints, thz variables, s k , that are con-
strainted to be nonnegative, sk 2 0,denote the tardiness in realizing event k,
and the corresponding penalty, pksk, appears in the objective function. Just as
before, this problem can be solved by an efficient network flow algorithm (see
Reference 5).
Another extension of the "CPM" problem was made by Moore13 and others.
They considered the determination of the optimal project schedule for the case
where multiple objectives were considered. For example, the completion time
of the entire project, as well as intermediate milestone events; duration of a par-
ticular set of activities that collectively should be allowed a specified minimum
time; and an attempt to operate within a futed budget for total direct project
costs. This generalized model for the time-cost crashing problem is solved using
goal programming, rather than linear programming. Modest sized networks
could be handled by this procedure.

Formulation for More Complex Time-Cost Functions

Chapter 8 discussed the use of a piece-wise linear approximation of a nonlinear

but convex activity time-cost trade-off function. Suppose, for example, the
trade-off curve for activity i-j is as shown in Figure 8-14 below, which is con-
tinuous, convex, and nonincreasing. The actual tradeoff curve is shown at the
left of Figure 8-14, with the two curves on the right depicting the separate seg-
ments of the piece-wise linear approximation to the actual curve. This repre-
sentation of yij in the objective function and the constraint equations is given
below, where only that portion of the formulation pertaining to activity i-j is
given.
Maximizef [Y] = . . . + C,ylij + C 2 ~ 2 +
i j. . . (64
Subject to

Using the simplex method to solve this problem will bring y,ij and yZij into the
solution in the proper order, i.e., y1# will remain its maximum value until y,ij is
reduced to zero, and then only will y,ij be reduced below its maximum value.
This follows because Cl > C, , and the sum Clylii + Czyzii is being maximized.
I t is obvious that this would prevail for any number of straight line segments, as
long as the actual trade-off is convex, which insures that Ci > Ci+,, for all i.
 Applications of Mathematical Programming 265

Figure 8-14 Breakdown of a continuous convex activity timecost trade-off curve.

Formulation for Continuous Nonconvex Activity

Time-Cost Trade-off Curves

If the trade-off curve is continuous, concave, and nonincreasing, the segments of

a piece-wise linear approximation to the actual curve can be brought into the
problem in their proper order by employing a nonnegative integer-valued vari-
able. An approximation of this type is shown in Figure 8-15 below, where the
actual cruve is shown on the left and the separate segments of this piece-wise
linear approximation are shown on the right. In this case, the representation
of yii in the objective function and the constraint equations is given below,
where again only that portion of the formulation pertaining to activity i-j is
given
Maximizef [Y] = . . . + CIylii + C2y2ii + ....
(74
Subject t o

Ti+dij+y,tI'+y2it- Tj5 0 (7b)

<=m-d
O~ylV (74
0<
=~,~~5D-rn (74
f4m - 4 5 ~ 1 ~ (7e)
~(D-~)LY,v (70
6 = a non-negative integer (7g)

If one puts constraints (7e) and (7f) together, one finds that this system of
constraints actually requires that S be equal to either zero or one, as shown in
equation (8).

If yii is equal to its maximum value, then 6 = 1 since y2ii = D - m, and because
of the integer constraint on 6, it must continue to equal 1 as long asy,ij > 0.
266 11 I Advanced Topics

A
.

4 Vz,,
d rn 0 Yu 0 m-d

Activity duration time

Figure 8-15 Breakdown of a continuous concave activity timecost trade-off curve.

Consequently, y , must,~ as it should, remain equal to its maximum value of

rn - d, as long as y,ij > 0. When y,ij = 0, 6 can also equal zero, and only then
can y,ii be less than rn - d. Thus, one can see that the use of the integer-valued
variable, 6, forces the activity segments to vary in a manner dictated by thephys-
ical problem that they represent.
It is a simple matter to extend this formulation to more than two linear seg-
ments by introducing an additional integer-valued variable for each additional
segment as shown in equation (9).

where in general the constraint, 0 5 yt 2 di- di-,, holds. Also, it is important

to note that this formulation of the problem will work on convex trade-off func-
tion as well as concave functions; however, it is not needed in the former case,
as shown by equation (6). Because this formulation can be extended to any
number of straight line segments, concave or convex, it follows that this integer-
variable formulation can be used on any trade-off curve that is continuous and
nonincreasing, such as the curve shown in Figure 8-16 which is neither convex
nor concave.

Activity duration time

Figure 8-16 Illustrative nonconvex, nonconcave trade-off function.
 Applications of Mathematical Programming 267

Formulation for Feasible Point Time-Cost Trade-Off Functions

It often happens that an activity can only be performed in a small number of dif-
ferent times, which gives rise to a small set of feasible time-cost points. For ex-
ample, consider an activity i-j that can only be performed at a normal time, Dii,
or a crash time, d i i In this formulation of the problem, one requires two* non-
negative integer-valued variables as follows:
yDii = 1 if the activity duration is Dii
= 0 if the activity duration is dii
(10)
ydii = 1 if the activity duration is dii
= 0 if the activity duration is Dii
Using these integer-valued variables, the linear programming formulation is as
follows:
Maximize f [Y]= . . - -CD yD..- Cd..ya + . . .
ii 11 I o (lla)
Subject to

Ti + DijyDii + diiydii - Ti 5 0 (1 lb)

~ ~=1
Y D +Ydii (11~)
y~ i , ydii
i
= non-negative integers
. (1 ld)

If the activity has k different feasible timecost points, the above formulation is
extended by introducing one non-negative integer-valued variable for each feasi-
ble time-cost point and requiring that the sum of all of the variables be equal t o
one.
The use of integer variables in this manner is, indeed, a very powerful tool
which could be used in a number of other ways. Consider, for example, the
criticism put forth by Fondah17 that the CPM procedure does not account for
the fact that activities are sometimes correlated in that a speed up of activity A
must be accompanied by a speed up in say activity B as well, because it is ac-
complished by the use of a special resource which acts in common t o both of
these activities. In this case, one could accomplish the requirement that neither
or both of the activities A and B are augmented by adding a pair of constraints

*One variable would suffice in this case, since y~ = 1 - yd . however, the formulation
ii v.
given in equation (11) is used because it suggests the generahation to more than two
feasible time-cost points.
268 11 I Advanced Topics

to equation (1 I), i.e., ydA - ydB 5 0 and ydB ydA 5 0, where ydA and yd,
-

are already required to be non-negative integers by constraints like (I ld). This

pair of constraints accomplishes the desired result because they require, in ef-
fect, that both variables must be equal to zero or both must be equal t o one.

Solving Problems with Arbitrary Trade-Off Functions

It is not too difficult t o combine the above formulations t o handle more compli-
cated trade-off functions which are combinations of discrete points and continu-
ous curves. The details of this procedure can be found in the report by Meyer
and Shaffert2 the three basic transition forms that must be handled are enumer-
ated in problem 3 at the end of Chapter 8.
The application of the linear programming formulations given by equations
(4), (6), (7), and (1 I), to the network given in Figure 8-11, but with more com-
plicated activity trade-off functions, is presented in problem 4. The solution of
this very small (five activities) problem requires 13 constraints, and the restric-
tion that two of the variables (one for the concave cost curve for activity 1-2,
and one for the discrete cost points for activity 3-4) be integer valued. To solve
even this small problem using the simplex method, together with an integer pro-
gramming algorithm, will require the use of a computer if the task is t o be ac-
complished in a reasonable time and at a reasonable cost.
Current computer hardware and integerlprogramming software is such that
these methods will handle networks of at most 100 activities. Other recent ap-
proaches to this problem have been more successful. For example, Panagio-
t a k o p o ~ l o s ' has
~ developed an efficient algorithm to determine optimal solu-
tions to the "CPW problem for the discrete case. That is, where the time-cost
alternatives for each activity are specified by a set of feasible t i e - c o s t pairs,
rather than continuous functions. The program for this algorithm is available
from the author. I t will handle networks well in excess of 100 activities.
A dynamic programming approach to the discrete case "CPM" problem was
developed by Crowston, and extended by Hindelang and Mutk9 Their approach
is called decision CPM (DCPM) networks. I t consists of conventional PERT/
CPM AND nodes as well as OR nodes. The latter nodes are followed by a set
of discrete time-cost activity performance alternatives. The computation time
of this approach is said to grow linearly with the number of activities in the net-
work. Thus it is potentially capable of handling very large networks.
In this appendix, the application of mathematical programming to the time-
cost trade-off problem was introduced along with logical extensions to the ob-
jective function and the activity time-cost trade-off relationships. Finally, pro-
cedures to solve these various problems formulations were cited.
THE PERT
STATISTICAL
APPROACH

In Chapter 1, PERT was described as being appropriate for scheduling and con-
trolling research and development type projects, or others comprised primarily
of activities whose actual duration times are subject t o considerable chance varia-
tion. It is because of this variability that for projects of this type, the time ele-
ment of project performance is usually of paramount importance. While the
deterministic CPM approach, as described in Chapters 3 and 4, isquite frequently
applied to programs of this type, the single estimate of the average activity per-
formance time which it employs completely ignores the chance element asso-
ciated with the conduct of the project activities. For example, an activity which
is expected to take 10 days t o perform, but might vary from 9 t o 11 days would
be treated no differently than an activity which is also expected to take 1 0 days
to perform, but might vary from 2 to 25 days. The advantage of the PERT sta-
tistical approach, originally developed by D. C. M a l c ~ l r n ~and
~ ~others,
' is that
it offers a method of dealing with this chance variation, making it possible t o
allow for it in the scheduling calculations and finally using it as a basis for com-
puting the probability (index) that the project, or key milestones in the project,
will be completed on or before their scheduled date(s).
This type of analysis could be very helpful in arriving at an acceptable project
plan for implementation, as shown in box (5) of Figure 1-4,which illustrated
the overall dynamic network-based planning and control procedure. For exam-
ple, probabilities of meeting a required schedule, of greater than say 0.75, might
signal a green light to go with the project plan. Values between 0.25 and 0.75
270 11 I Advanced Topics

would signal a yellow caution light; careful attention to progress on this project
is appropriate, and some project replanning may be anticipated. Finally, a prob-
ability less than 0.25 would signal red to immediately stop and replan to achieve
a better chance of meeting the required schedule.
This type of early warning system has considerable merit when large uncer-
tainty in activity duration time exists. For example, the major overhaul of air-
craft involves considerable uncertainty that is not revealed until covers are re-
moved from control cables, engines, etc., where corrosion and wear may be found
to vary from minimal to extensive. In other projects, external elements, such as
the weather, may interject uncertainty. For example, in the Alaskan pipeline
project, risk analysis of the construction schedule required the inclusion of the
effects of weather conditions within the project network plan.

OVERVIEW O F PERT

To give an overview of the PERT statistical approach, consider the network

originally presented in Figure 4-5, and shown here in modified form in Figure
9-1. In this network, the critical path consists of the three activities 0-3, 3-7,
and 7-8. Now suppose that the performance of each of these activities is subject
to a considerable number of chance sources of variation such as the weather,
equipment failures, personnel or materials problems, or uncertainties in the
methods or procedures to be used in carrying out the activity. I t may be argued
that if difficulties of one sort or another are encountered on a particular activity,
that additional resources will immediately be applied t o this activity, or subse-

/ 7. t-
; - -
t,; :t

Figure 9-1 Illustration of the "conventional" PERT statistical approach to the network
originally presented in Figure 4-5.
 The PERT Statistical Approach 271

quent activities on the critical path, so that the project will still be completed
on time. This chapter is concerned with the problem of estimating the prob-
ability of having to undertake such measures. Hence, the uncertainties in per-
formance time being referred to here are those associated with completing the
originally defined activity with the originally specified resources.
Returning to the example, one notes that the actual performance times for
the activities on the critical path, instead of being exactly 2, 8, and 5 days, are
variables subject to random or chance variation, withmean values of 2, 8, and 5,
respectively. Also, the actual time t o perform the activities on the critical path
is the sum of three random or chance variables, and (except for the slight possi-
bility that activity 6-7 may be completed after activity 3-7 and activity 5-8
after activity 7-8) this sum is also the actual t i e to complete the project.
Hence, to estimate the statistical distribution of project performance time,
and in turn compute the probability of meeting a scheduled date for the comple-
tion of the project, it will be necessary to deal with the statistics of the sum of
random variables. In the next section the theory of probability and statistics
necessary to handle this problem will be considered. The following section will
describe "conventional" PERT scheduling and probability calculation along with
several practical applications. Several important refinements t o the conventional
PERT procedure are given at the end of this chapter.
The treatment of PERT given in this chapter is referred t o as "conventional"
because the calculation of earliest and latest activity start and finish times is
made in the same way as described in Chapter 4, using expected activity perfor-
mance times only. The variability associated with activity performance times is
involved only in the computation of PERT probabilities. This simplifying as-
sumption causes the PERT event times t o be biased slightly, always on the low
side, and the PERT probabilities considerably on the high side. For example,
in Figure 9-1 the earliest expected complete time for activity 7-8 is 15. How-
ever, since the actual complete time for activity 5-8 may be greater than that for
activity 7-8, even though its expected time of 14 is less (earlier) than that for
activity 7-8, the correct expected time for the completion of both activities 5-8
and 7-8 must be greater than 15.
The fact that the conventional PERT procedure always underestimates the
expected occurrence times of network merge events and their successors, dis-
turbs users. The fact is, however, that this bias is usually relatively small; less
than 5 percent. It turns out that the PERT probability estimates are more seri-
ously biased, on the high side. This is due not only to the merge event t i e bias,
but also to the relatively large error in the estimate of the variance of the actual
project duration time. These biases will be illustrated in the latter sections of
this chapter, along with two practical methods of circumventing them. One rela-
tively simple method, called PNET, considers not only the critical path, but also
a "sufficient" set of near critical paths so that accurate probability estimates
272 11 I Advanmd Topi-

can be made. The other method is Monte Carlo simulation. It is an order of

magnitude more costly than PNET, but has the very important added advantage
of estimating the criticality of all network activities and events. The latter is
expressed as a probability that any activity in question will end up on the critical
path when the project is actually conducted.
The PERT statistical approach to project management is not widely used to-
day, even though its potential has grown over the past decade due to the pro-
liferation of very costly and time consuming large scale systems projects. Cran-
da114 has reported that the deterministic activity performance time approach has
proven costly and invalid in projects such as the fabrication of offshore oil plat-
forms and storage tanks, and projects in Alaska where weather is a factor in con-
struction and logistic supply. Many other projects of this type could be cited,
particularly large government hardware development projects whose history of
large uncertainty in time and cost performance is well known. In view of this,
it is hard t o say why PERT is rarely applied. Certain contributing factors are
known. The lack of validity of the basic assumptions underlying PERT have
been cited in scores of technical papers. But by and large, most of these prob-
lems could be affordably circumvented by the two methods cited above. This
reduces t o the conjecture that top level project managers do not understand the
basic principles of probability and statistics, or given that they do, they have just
not learned how to use PERT to solve their management problems. For this
reason, the next section is devoted to basic probability theory. While many
readers of this text will already have a background in this material, they may
find this material interesting as an example of how these concepts might be
explained to managers who are to use the PERT probability approach to project
planning and control. The material is presented with this objective in mind.

BASIC PROBABILITY THEORY

Probability as a Measure of Uncertainty

Some of the mathematical definitions of probability become highly abstract and

the language somewhat complicated, although they will be discussed in less for-
midable language below. In a sense, however, one already knows what prob-
ability is all about. If one is told that an event is "almost certain," "highly prob-
able," "about fifty-fifty," "highly unlikely," or "highly improbable," one has a
good intuitive feel for the meaning of what is being said and, furthermore, this
intuitive feeling is correct. All that probability theory attempts to dois to quan-
tify these somewhat subjective statements in a precise and objective way.
In order to do this it has been found convenient to express probabilities on a
scale that runs from 0 to 1. On this scale, zero represents impossibility and one
represents certainty; the numbers in between represent varying degrees of
likelihood. Instead of saying, for example, that it is "almost certain" that a de-
 The PERT Statistical Approach 273

vice will continue operating for at least one hour and "highly improbable" that
it will continue operating for more than one thousand hours, one can say that
the respective probabilities are, say, 0.999 and 0.001. The definitionsandmathe-
matical procedures that enable one t o go from qualitative to quantitative state-
ments can become quite technical and highly specialized; the intent, however, is
to enable one to make precise and valid statements about the degree of certainty
or uncertainty associated with specific occurrences.

The Managerial Function: Decision Making Under Uncertainty and Risk

The words "uncertainty" and "risk"? appear frequently in mathematical litera-

ture on probability concepts. The same two words or their synonyms are also a
part of management's vocabulary, for the prime function of management is de-
cision making under conditions of uncertainty with the objective of balancing
the risks associated with a particular problem. Risk itself has two elements: the
probability that something will happen, and the loss that will result if it does
happen.
Consider the trivial example of deciding whether or not to wear a raincoat to
work. If one decides to take a raincoat and it does not rain, there will be a loss,
the effort or nuisance involved in carrying the raincoat. If one does not take the
raincoat and it does rain; another kind of loss is involved, i.e., getting wet. The
decision will, therefore, depend upon an evaluation of these possible losses and
an assessment of the probability of rain. Current industrial applications of prob-
ability theory revolve around this concept of balancing risks. Probability theory
is used, for example, in the determination of optimal inventory sizes, where the
opposing risks are the costs of carrying too much stock, and the loss of sales that
results when an out-of-stock condition occurs.
Probability has already been defined as a way of measuring uncertainty. Inas
much as the problems facing management are the problems of uncertainty and
risk, it is clear that probability has an important role to play in helping the man-
ager to formulate and solve these problems.

Empirical Frequency Distributions

To present a background in probability and statistics it is logical to begin by con-

sidering the basic raw material of statistics, i.e., observations of some measurable
quantity subject to random or chance sources of variation. For discussion pur-
poses consider a PERT activity which has been performed in the past a large
number of times under essentially the same conditions. This assumes that no

?The definition of these two terms used here differs from that used in literature on decision
theory, where the differentiation between uncertainty and risk is based on the presence or
absence of a probability distribution associated with the variable in question.
 Activity duration time, f
Figure 9-2 Empirical frequency distribution of activity duration times.

learning, changes in working conditions, job description, etc., take place. Al-
though PERT generally involves no statistical sampling of this sort, for the
purpose of this discussion, one may suppose the duration times for this ac-
tivity ranged from 7 to 17 days. Now suppose that one counts the number of
times the activity required 7 days to perform, 8 days to perform, etc., and dis-
plays the resulting data in the form of an empirical frequency distribution or
histogram as shown in Figure 9-2. If one had an infmite number of observa-
tions and made the wjdth of the intends in Figure 9-2 approach zero, the
distribution would merge into some smooth curve; this type of curve will be
referred to as the theoretical probability density of the random variable. The
total area under such a curve is made to be exactly one, so that the area under
the curve between any two values o f t is directly the probability that the random
variable t will fall in this interval.

Characterization of an Empirical Distribution

To describe an empirical frequency distribution quantitatively, two measures

are frequently employed-one which locates the point about which the distribu-
tion is centered, a measure of its central tendency or location, and the other
which indicates the spread or dispersion in the distribution, a measure of its
variability. These measures are illustrated in Figure 9-3. At the top and middle
of the figure, the two distributions differ either in their mean values or disper-
sion, while at the bottom they differ in both respects. This same information
is given by quantitative measures of these two characteristics of a frequency
distribution.
In PERT computations, this text will use the familiar arithmetic average or
 The PERT Statistical Appmach 275

!j Two distributions with same dispersion but different means

E

g Two distributions with same mean but different dispersion

C
'y

!j Two distributians with different meansand dispersion

C

2I
-8
D
$
m

-
u
E
>
u
,-
-
u
a3
a:
Random variable

Figure 9-3 Illustration of differencw in measures of central tendency and dispersion.

mean as a measure of central tendency, and what is called the standard devia-
tion as the measure of variability. These statistics will first be defmed with re-
spect to a sample of n observations drawn from some distribution such as the
one shown in Figure 9-2. If the n obselvations are denoted by t,, t , ,. . , t,, .
these measures are computed as follows:

Measure of central tendency = arithmetic mean

=(tl + t 2 + . . - + tn)/n = i (1)
276 11 I Advanced Topics

Measure of variability = standard deviation = st

The above formula for the standard deviation indicates why it is sometimes
referred to as the root-mean-square deviation; it is the square root of the mean
of the squares of the deviations of the individual observations from their aver-
age. Computations will frequently use the square of the standard deviation,
which, for convenience, is called the variance; sf = variance oft.

Physical Interpretation of the Mean and Standard Deviation

The question usually asked at this point is what do I and s, (or st) mean? First
of all, t and st both carry the same time units and are estimates of the true mean
and standard deviation of the distribution shown by the smooth curve in Figure
9-2. These quantities will be denoted by t, and (vt)ll*, respectively; I ap-
proaches t, and st approaches ( v ~ ) 'as~ the
~ size of the sample, n, approaches
infinity. If some assumption is made now about the theoretical distribution (the
smooth curve in Figure 9-2) from which the sample was obtained, one can pro-
ceed with the interpretation. For example, suppose the random variable t is
"normally" distributed, that is, the distribution has a characteristic symmetrical
bell shape which frequently occurs when a variable is acted upon by a multitude
of random chance causes of variation. In this case, our interpretation is shown
in Figure 9-4.

Central Limit Theorem

The last bit of statistical machinery needed for PERT probability computations
is the Central L i t Theorem, which is perhaps the most important theorem in
all of mathematical statistics. In the context of PERT, this theorem may be
stated in the following way.

Central Limit Theorem

Suppose m independent tasks are to be performed in order; (one might think

of these as the m tasks which lie on the critical path of a network). Let tl , t , ,
.
. . , tm be the times actually required to complete these tasks.
.
Note that these are random variables with true means t,, , t,,, . . , t,, and
true variances V, , Vt2,. . . , Vm, and these actual times are unknown until
these specific tasks are actually performed. Now define T t o be the sum:
T=t,tt,+..-tt,
and note that T is also a random variable and thus has a distribution. The Cen-
tral Limit Theorem states that if m is large, say four or more, the distribution of
 The PERT Statistical Approach 277

Scale of
random
variable f

re - 3(V,In 95% of area

I
t, + ~ I v , ) "
\ 1
v
99.7% ofaarea
Figure 9-4 Selected areas under the normal distribution curve.

T is approximately normal with mean E and varianw VT given by

That is, the mean of the sum, is the sum of the means; the variance of the sum
is the sum of the variances; and the distribution of the sum of activity times will
be normal regardless of the shape of the distribution of actual activity perfor-
mance times (such as given in Figure 9-2).
The normal distribution is extensively tabulated and therefore probability
statements can be made regarding the random variable T by using these tables.
A table of normal curve areas is given in Appendix 9-1, and an example illus-
trating its use is given later.

The Dice Tossing Experiment

To establish confidence in, and further understanding of the Central Limit

Theorem, it is worthwhile to study its application to a familiar experiment-
dice tossing. An experiment of tossing a single die can be described as shown
in Figure 9-5.
278 II I Advanced Topica

Experiment: Tossing a single die

Possible outcomes:

Random variable X 1 2 3 4 5 6
Figurs 9 5

If the die being tossed is unbiased, the probability of each outcome of this
experiment is equally likely, and hence has a probability of 8,
since there are
just six possible outcomes. This is shown in Figure 9-6, which is called a theo-
retical probability distribution for the random variable X. The mean and vari-
ance of this theoretical distribution can be computed using equations (1) and
(2) in which each possible value of the random variable X is weighted by its
theoretical probability. These results are indicated in Figure 9-6.
MeanofX=(l x $ ) + ( 2 X i ) + ( 3 ~ i ) + ( 4 ~$1
+(5~;)+(6xi)=$=3;
Variance of X = (1 - 3'); X i+ (2 - 3+)2 X i
t(3-332 X i + ( 4 - 332 x i
+(5-3f)2~$+(6-3;)2~$
= $=211
12

Now consider tossing two dice with the random variable Y defined as the sum
of the spots on both dice, i.e., Y = X, + X2.The probability distribution for
this example, shown in Figure 9-6, follows directly from the fact that there are
36 mutually exclusive and equally likely ways of tossing two unbiased dice, the
outcomes of which are shown by the matrix in Figure 9-7, which gives the total
number of spots for each of the 36 combinations.
The reader can verify that by using equations (1) and (2) in the same manner
as shown above for the single die case, the mean and variance of this distribution
are exactly twice the values for a single die, i.e., mean = 2 X 34 = 7 and the
variance = 2 X 2g = 56, as they should be according to the Central Limit Theo-
rem. One should also note from Figure 9-6 that while the basic random variable
X has a rectangular distribution, the random variable Y has a triangular distribu-
tion which represents a large step toward the theoretical normal distribution as
dictated by the Central Limit Theorem.
To carry this experiment still one step further, consider tossing three dice
with the random variable Z defined as the sum of the spots on all three dice,
 The PERT Statistical Approach 279

Single die Two dice

MwnofX=3112 Meanof Y = 7
VarianceofX= 2 11/12 P IYI Variance of Y = 5 516

P (XI

Number of spots on a single dieX Total number of spots on tw dice = Y

P fZl
I Three dice
Mean of Z = 10 1 R
Variance of Z = 8 314

Total number of spots on three dice = Z

Figure 9-6 Probability distribution for tossing unbiased dice.

i.e., Z = X, + X2+ X 3 . The reader can again verify that the mean and variance
of this distribution follow the Central Limit Theorem, i.e.,
Mean of Z = 3 X 3 1 = 10;
Variance of Z = 3 X 2% = :8
Examination of Figure 9 6 indicates that the shape of the distribution is now
very close to the theoretical normal distribution. The Central Limit Theorem
will be applied in the section below on the probability of meeting a scheduled
280 11 I Advanced Topi-

Second die

1 2 3 4 5 6

Figure 9-7 The thirw-six equiprobable ways of tossing two unbiased dice.

date, while an empirical verification of this theorem is taken up in exercise 1

at the end of this chapter.

PERT SYSTEM OF THREE TIME ESTIMATES

PERT basic scheduling computations utilize the expected values, t,, of the hypo-
thetical distributions of actual activity performance times, as depicted in Figure
9-2 and 9-8 below. Since PERT addresses itself primarily to programs whose
activities are subject to considerable random variation and to programs where
time schedules are of the essence, it utilizes the standard deviations of the dis-
tributions shown in these figures in computing a measure of the chances of meet-
ing scheduled dates of project milestones.
In making PERT computations, it must be realized that the activity perfor-
mance time distribution is purely hypothetical, since one ordinarily is unable
t o do any statistical sampling whatever. [If historical (sample) activity duration
data are available, they can be used to estimate a, rn, and b, as indicated in Ap-
pendix 9-21. After an activity has been performed, the observed actual perfor-
mance time, denoted by t, for the activity can be considered as a single sample
from this hypothetical distribution. However, all computations are made prior
to the performance of the activity; hence, as stated above, the basis of PERT
computations involves no statistical sampling, but rather it depends on the judg-
ment of the person in charge of the activity in question. The latter judgement
is, of course, based on a sampling of work experiences; however, this is not
sampling in the strict statistical sense. In making these subjective estimates,
 The PERT Statistical Approach 281

Probability
density

& = $= standard deviation

Optimistic Most likely time Pessimistic time

time
Figure 9-8 PERT system of three time estimates.

one is asked to call on his general experience, and his knowledge of the require-
ments of the activity in question, to consider the personnel and facilities avail-
able to him, and then to estimate the three times shown in Figure 9-8. These
times defmed below, will then be used to estimate the mean and standard devia-
tion of the hypothetical activity performance time distribution. The choice of
a most likely time, rn, and then a range of times from an optimistic estimate, a,
to a pessimistic estimate, b, seems to be a natural choice of times.

Definition:

a = optimistic performance time; the time which would be bettered only one
time in twenty if the activity could be performed repeatedly under the same es-
sential conditions.

Definition:

rn =most likely time; the modal value of the distribution, or the value which
is likely to occur more often than any other value.

Definition:

6 = pessimistic performance time; the time which would be exceeded only

one time in twenty if the activity could be performed repeatedly under the same
essential conditions.
282 11 I Advanced Topics

The above definitions of a and b are called the 5 and 95 percentiles, respec-
tively, of the distribution of the performance time, t. These definitions are
based on a study by Moder and Rodgers.lD They differ from the original devel-
opment of PERT,^^" where they were assumed to be the ultimate limits, or the
0 and 100 percentiles of the distribution of t. An intuitive argument for our
definition is that since the estimates of a and b are based on past experience and
judgment, the 0 and 100 percentiles would be very difficult to estimate, since
they would never have been experienced. Further arguments of a statistical na-
ture will be presented below.
Although the above definitions of a, m, and b appear to be clear and work-
able, the following points will be helpful in obtaining reliable values for these
time estimates.
1. One of the important assumptions in the Central L i i t Theorem is the in-
dependence of the random variables in question. Since this theorem is the
basis of PERT probability computations, the estimates of a, m, and b
should be obtained so that the assumption of independence is satisfied,
that is, they should be made independently of what may occur in other ac-
tivities in the project, which may in turn affect the availability of man-
power and equipment planned for the activity in question. The estimator
should submit values for a, m ,and b whiih are appropriate if the work is
carried out with the initially assumed manpower and facilities, and under
the assumed working conditions.
2. The estimates of a, m, and b should not be influenced by the time avail-
able to complete the project, i.e., it is not logical to revise estimates by an
across-the-board cut in t i e s after learning that the project critical path is
too long. This completely invalidates the PERT probabilities and destroys
any positive contribution that they may be able to make in the planning
function. Time estimates should be revised only when the scope of the
activity is changed, or when the manpower and facilities assigned to it are
changed.
3. To maintain an atmosphere conducive to obtaining unbiased estimates of
a, m, and b, it should be made clear that these are estimates and not
schedule commitments in the usual sense.
4. In general, the estimates of a, m, and b should not include allowances for
events which occur so infrequently that one does not ordinarly think of
them as random variables. The estimates of a, m, and b should not include
allowances for acts of nature-fires, floods, hurricanes, etc.
5 . In general, the estimates of a, m, and b should include allowances for
events normally classed as random variables. An example here would be
the effects of weather. For activities whose performance is subject to
weather conditions, it is appropriate to anticipate the time of the year
when the activity will be performed and make suitable allowance for the
anticipated prevailing weather in estimating a, m,and b.
 The PERT Statistical Approach 283

Estimation of the Mean and Variance of the Activity Performance Times

It is commonly known in statistics that for unimodal distributions the standard

deviation can be estimated roughly as $ of the range of the distribution. This
follows from the fact that at least 89 percent of any distribution lies within
three standard deviations of the mean, and for the normal distribution this per-
centage is 99.7+ percent. Hence, one can use time estimates, a and b, to esti-
mate the standard deviation (vt)',)'I2 or the variance, V,, as shown in equation(3):

(vt)'I2 = (b - a)/3.2, or Vt = [(b- a)/3.2] (3)

As mentioned above, a and b were originally defined as the 0 and 100 percen-
tiles of the distribution of t , and therefore, the divisor in equation (3) was 6 in
place of the above value of 3.2, in the original development of This
is the basis of another argument in favor of our 5 and 95 percentile definitions
of a and b. In the paper by Moder and ~ o d ~ e r s , ' it
' is shown that the difference
(b - a) varies from 3.1 to 3.3 (average of 3.2) standard deviations for a wide vari-
ety of distribution types ranging from the exponential distribution to the normal
distribution, including rectangular, triangular, and beta type distributions. For
this same set of distributions, the difference between the 0 and 100 percentiles
varies, however, from 3.5 all the way to 6.0. Thus, the use of the 5 and 95 per-
centiles for a and b leads to an estimator of the standard deviation that is robust
to variations in the shape of the distribution of t. This is of some importance,
because in general we do not know the shape of the distribution o f t , and fur-
ther, we wish to avoid making any specific assumptions about it.
A simple formula for estimating the mean, t,, of the activity time distribution
has also been developed. It is the simple weighted average of the estimates a, m,
and b given in equation (4).

Mean = t, = (a + 4m + b)/6 (4)

To derive this formula for the mean, one must assume some functional form
for the unknown distribution o f t , such as shown in Figure 9-8. A likely candi-
date, chosen by lark,^ is the well known beta distribution, which has the de-
sirable properties of being contained inside a finite interval, and can be sym-
metric or skew, depending on the location of the mode, m, relative t o a and b.
Lacking an empirical basis for choosing a specific distribution, the beta distribu-
tion was historically accepted as a mathematical model for activity duration
times, for purposes of deriving equation (4) only. Using this distribution as a
model and assuming that equation (3) holds,* then t, is a cubic polynomial in
m. Equation (4) is a linear approximation to the exact formula, whose accuracy
is well within limits dictated by the accuracy of the estimates of a , m , and b.

*The 0 and 100 percentile definitions of a and b were used in this derivation. The same
formula should hold, however, for the 5 and 95 percentile definitions used in thjs text.
284 11 / Advanced Topia

It should also be pointed out that the mean is equal to the most likely or
modal time ( t , = m ) , only if the optimistic and pessimistic times are symmetri-
cally placed about the most likely time, i.e., only if b - m = m - a. Thus, in
the CPM procedure the single time estimate, denoted by D, is an estimate of the
mean activity duration time and is not necessarily the most likely time as de-
fined here. This is essential, since according to the Central Limit Theorem, the
expected total duration of a series of activities is the sum of their mean times
and not a sum of their most likely times. In fact, since the distribution of the
sum of random variables tends to the normal (symmetrical) distribution for
which the mean and the mode are the same, the most likely (modal) duration
time of a series of activities is not given by the sum of the individual activity
most likely times, but rather by a sum of their mean times.
If a single time estimate system is being used, and an activity is encountered
which has a skew distribution with a considerable amount of variation, then
equation (4) might be of assistance in arriving at the single time estimate, D. In
this case, a person might feel that he can estimate the mean activity duration
time more accurately by estimating a, m , and b, and using equation (4) to con-
vert these numbers to the required single t i e estimate, D. Some evidence t o
this effect was found in the study cited in Reference 10.
To illustrate the computation in the PERT statistical approach, consider the
simple network given in Figure 9-9. Here, for the network originally presented
in Figure 4-1, single time estimates have been replaced by estimates of a, m , and
b. For example, a = 1, m = 2, and b = 3 for activity 1-2. In the middle diagram
of Figure 9-9 are indicated the values of t,, and Vt computed from equations
(3) and (4). These computations are illustrated below for activity 1-2.
t, = (1 + 4 X 2 + 3)/6 = 2
V, = [(3 - 1)/3.2] e0.391
The results of the forward pass are indicated by the time scale at the bottom
of Figure 9-9. The earliest expected event occurrence times, E, are computed in
exactly the same manner (in conventional PERT) as outlined for the single time
estimate systems, as shown in Figure 4-1.
The event variance, VT, is computed in a manner quite similar to the compu-
tation of E. The rules are as follows:

R U LE 1. VT for the initial network event is assumed t o be zero.

RULE 2. The VT for the event following the activity in question in obtained by
adding the activity's variance, V*,to the variance of the predecessor
event, except at merge events.
RULE 3. At merge events, VT, is computed along the same path used to obtain
E, i.e., the longest path. In case of ties, choose the path which gives
the larger variance.
 The PERT Statistical Appmach 285

Basic network with three time estimates, a, m, and 6.

Basic network with t, and V , computed for each activity

Mean = E6 = 12
3 Std. dew. = 3 ( V 7

Time units
Figure 9-9 PERT statistical computations.
286 11 / Advanced Topica

Applying these rules to the network given in Figure 9-9, one obtains the fol-
lowing:
VT (event 1) = 0

VT (event 3) = 0.391 + 1.562 = 1.953

VT (event 4) = 0.391 + 0.391 = 0.782
Since event 5 is a merge event, its variance according to Rule 3 above is com-
puted along the path 1-2-3- 5.
VT (event 5) = 1.953 + .391 = 2.344
VT (event 6) = 2.344 + .391 = 2.735
The backward pass and slack computations are also performed in the same
manner shown in Figure 4-1; they are not given in Figure 9-9.

PROBABILITY OF MEETING A SCHEDULED DATE

Although scheduled dates could be applied to the start or finish of a project

activity, they have traditionally been applied to the time of occurrence of net-
work events. Scheduled dates are usually specified only for those events that
mark a significant state in the project and vitally affect subsequent project ac;.
tivities; such events are frequently called milestones. In this section, the prob-
lem of computing the probability of occurrence of an event, on or before a
scheduled time, is considered.
Refemng to Figure 9-9, the critical path for this network can be seen to be
1-2-3-5-6. Now consider the time to perform each of the activities along this
path as independent random variables, the same assumption made during the
process of collecting the a, m, and b activity time estimates. Furthermore,
the sum of these random variables, which shall be denoted by T, is itself a
random variable which is governed by the Central Limit Theorem. Therefore

Mean of T = E6 = (t,),-, + (t,),, + (te)3-5 + (te)S-6

E,=2+4+3+3=12
Variance of T = V T = Vtl-, + Vt2-, + Vt3+ + vtS-6
VT = 0.391 + 1.562 + 0.391 + 0.391 = 2.735
and finally, the Central L i t Theorem enables one to assume that the shape of
the distribution of T is approximately normal. This information is summarized
 The PERT Statistical Approach 287

Mean = E8 = 12

Probability ( T < TS = 14) Standard deviation =

T
7 12 14 17

Figure 9-10 Distribution of the actual occurrence time (T) of event 6.

in Figure 9-10, where the distribution is shown to "touch down" on the abscissa
at three standard deviations on either side of the mean, i.e., at 12 f 5.
Now, the problem of computing the probability of meeting an arbitrary
scheduled date, such as 14 shown in Figure 9-10, is quite simple. Since the total
area under the normal curve is exactly one, the crosshatched area under the nor-
mal curve is directly the probability that the actual event occurrence time, T ,
will be equal to or less than 14, which is the probability that the scheduled date
will be met. This probability can be read from the table of normal curve areas,
given in Appendix 9-1 at the end of this chapter. In order for this table to apply
to any normal curve, it is based on the deviation of the scheduled date in ques-
tion, Ts, from the mean of the distribution, E6, in units of standard deviations,
( v T ) " ~ . Calling this value Z , one obtains

A value of Z = 1.21 indicates that the scheduled time, T8, is 1.21 standard de-
viations greater than the expected time, E, = 12. Reference to Appendix 9-1B
indicates that this value of Z corresponds to a probability of approximately
0.89. Thus, assuming that "time now" is zero, one may expect this project to
end at time 12, and the probability that it wiU end on or before the scheduled
time of 14, without expediting the project, is approximately 0.89. It should be
pointed out that if Ts had been two days less than E, instead of being greater,
i.e., Ts = 12 - 2 = 10, then Z = - 1.21, and the corresponding probability would
be 0.11. Hence, it is essential that the correct sign is placed on the 2.
The above phrase, "without expediting," is very important. In certain proj-
ects, schedules always may be met by some means or another, for example, by
changing the schedule, by changing the project requirements, by adding addi-
tional personnel or facilities, etc. The probability being computed here is the
probability that the original schedule will be met without having to expedite the
work in some way or another. For this reason, the following rules should be
adopted in dealing with networks having two or more scheduled dates.

Definition:

The probability of meeting a scheduled date is the probability of occurrence

of an event on or before a specified date (time).

Rule:

To compute the probability of meeting a scheduled date, the variance of the

initial project event should be set equal to zero, and all scheduled dates other
than the one being considered should be ignored in making the variance and
probability computations.

Definition:
The conditional probability of meeting a scheduled date is the probability
of the occurrence of an event on or before a specified time, assuming that all
prior scheduled events occur on their scheduled dates.

Rule:

To compute the conditional probability of meeting a schedule date, set the

variances of the initial project event and all scheduled events equal to zero, and
then make the usual variance and probability computations.
The above definitions and rules suggest where each of these two probabilities
might be applied. If one is concerned primarily with the planning of a subnet-
work consisting of the activities between two scheduled events, then the condi-
tional probability is pertinent. However, if one is concerned with the entire
project, then the unconditional probability of meeting a scheduled date seems
pertinent, since it gives the probability of having to expedite a project sorne-
where in order to meet each of the scheduled event times.

COMPUTATION OF CUMULATIVE PROBABILITY

VS. PROJECT DURATION

Suppose the scheduled time for the completion of the project shown in Figure
9-9 had been T, = 10; then the probability of meeting this schedule would have
 The PERT Statistical Approach 289

Table 9-1. Computation of Cumulative Probability

of Meeting Alternative Project Scheduled Times

Ts 2 = IT, - 12111.654 P{T 5 T,}

8 -2.43 0.008
10 -1.22 0.1 11
12 0 0.500
14 1.22 0.889
16 2.43 0.992
18 3.65 1.000

been the relatively low figure of 0.11, as discussed in the previous paragraph.
One might then ask the question, what is the probability of meeting various re-
vised project completion times? A convenient way to answer this question is
with a graph (or table) giving cumulative probability vs. project duration time.
Applying equation (5) above, this probability can be written as

Cumulative Probability
Completing Project on = P{T 5 T,} = P{Z $ [T, - E(T)]1 6 )
Before Time T,

Treating T, as a variable, a cumulative probability curve, such as shown in Table

9-1 and Figure 9-11, can be obtained from the following expression, together
with Appendix 9-1.

P{T 5 T,} = P{Z = [(T, - 12)/1.654])

Project scheduled duration time

Figure 9-11 Comulative probability of completing the project on or before the scheduled
time TS.
 290 11 I Adnncsd Topics

Now, the probability of meeting any possible schedule can be read directly from
Figure 9-1 1. For example, the probability of meeting a schedule of T, = 13 days
is approximately 0.7.

ILLUSTRATIVE EXAMPLE USING PERCENTILE ESTIMATES

To illustrate the use of Eqs. (3) and (4), an example wiU be given dealing with
the overhaul of jet aircraft. The data given in Table 9-2 lists the nine activities
that form the critical (longest) path through the network of overhaul activities.
Because of physical space limitations, the number of men that can be assigned
to an overhaul task is almost fixed. Thus, an activity cannot be accelerated by
adding manpower when an overhaul project is behind schedule, and for this rea-
son, it is reasonable to assume that the actual activity duration times are statis-
tically independent. Also, the critical path is considerably longer than the next
most critical path, so that the usual PERT application of the Central Limit
Theorem is justified. The results given in Table 9-2 indicate that the expected
duration of this series of activities is 169.0 hours (7.05 days), with a standard
deviation of the actual performance times equal to 11.3 hours (0.47 days). The
overhaul duration that will not be exceeded more often than say one time in ten
can be estimated by substituting these statistics in eq. (5) and solving for Ts,
which is found to be 7.65 days.
In this example, the objective was to accomplish the overhaul project in seven
days. Thus, technological changes in the techniques of overhaul or in the extent
of the work to be accomplished must be made to reduce the overall duration of

Table 9-2. P E R T Probability Analysis o f an Aircraft Overhaul Project

Estimated Activity Durations (Hours)

Opti- Most Pessi-
Activjtv Description mistic Likely mistic Meanlime" Variance"

Open Pylons 4
Open Engines 1
Cable Checking 10
Remove Engines 1
Pylon Rework 96
Reassemble Pylon, Etc. 12
Fuel Aircraft 1
Check Fuel Tanks, Etc. 4
Wing Closures 2
Final Checkout 6
TOTALS: ..... ... . . .... ....... . . . ..... . ...... .. ...
Standard Deviation = 11.3 Hours

* Based on Eq. (3)

Based on Eq. ( 4 )
 The PERT Statistical Approach 291

the project by at least 0.65 days, or about two work shifts, or a reduction in the
standard deviation of the project duration, or some combination of both.

MONITORING ACTIVITY TIME ESTIMATES AND PERFORMANCE

One of the frequent criticisms of the PERT statistical approach is that the per-
sons supplying values of a, m , and b do not have the experience to furnish accu-
rate data. This criticism is largely due to the nature of the work being planned.
Also, the estimators are not consistent, some being conservative while others are
liberal in making their estimates. In addition, there is the real possibility that
the estimates are biased by a knowledge of what some h~gherauthority would
like the times to be for aribtrary reasons.
This problem was studied by MacCrimmon and Ryavec.' They studied the
effects of various sources of errors on the estimates of the mean, t,, and the
standard deviation (v,)"~. They considered errors introduced by (1) assuming
the activity time distribution was a beta distribution; (2) by using the PERT ap-
proximate formulas given by equations (3)* and (4); and (3) by using estimates
of a, m, and b in place of the true values. They concluded that these sources of
error could cause absolute errors in estimates of t , and ( ~ , ) ' ~ % f 3 0 and 15
percent of the range (b - a), respectively. Since these errors are both positive
and negative, however, they will tend to cancel each other.
This problem was also considered by King and Wilson,' who studied actual
data obtained from a large scale development project involving a prime con-
tractor and a number of subcontractors. They examined the hypothesis that
there is a general increase in the accuracy of pre-activity time estimates as the
beginning of the activity approaches. They rejected this hypothesis. Their
data also indicated that most of the time estimates were optimistic, some being
as low as 13 percent of the actual activity duration time. The same conclusion
was reached in regard to ability t o improve the estimates of remaininglife after
the activity was started. In this study, the estimate of remaining life was, on
the average, 72 percent of the actual value. On the basis of these findings, the
authors proposed for consideration, the upward adjustment of all time estimates.
In the project studied, a multiplier of 1.39 would have reduced the average error
in the activity duration times to zero.
Another serious aspect of this problem that has not received much attention
is the effect of the "Peter Principle" applied t o activity performance times. It
might be stated as follows: The actual time required to perfoim an activity wilI
rise to at least fill the time allotted to it. Referring to Fig. 9-9, the actual time to
perform an activity would be expected t o be less than the mean, t,, about as

*They assumed 0 and 100 percentiles were used, and a corresponding divisor of 6 in equa-
tion (3).
292 11 I Advanced Topics

often as it exceeds the mean. (About is used here because for a distribution
skewed to the right (or left), the mean will be slightly greater (or less) than the
median or 50% point.) Experience would undoubtedly not bear this out be-
cause of the ever present human temptation t o take a situation where the per-
formance time could be less than the mean, and drag it out to refme or improve
the results beyond the initial product specification, or by the diversion of efforts
away from the activity to complete "other things." Thii presents a real chal-
lenge to management to create a climate where performance times less than the
expected values (t,) will occur. The control chart procedure described below
may be of some help in alleviating this problem.
Another approach to this problem, which these authors feel has more promise,
is t o work with the individuals making the time estimates t o improve their future
estimates by supplying them with positive feedback information. It is suggested
that records be kept for each person supplying activity time estimates. These
records should give the deviation of the estimated and actual activity perfor-
mance times in units of standard deviations as shown in equation (6).

Z = difference between estimated and actual duration time divided

by the estimated standard deviation of the duration time

One problem which complicates this analysis is that the specifications for the
work comprising the activity, or the level of effort applied to the activity, may
be changed before the activity is completed. In these cases, the only valid proce-
dure is t o use the estimated mean time, t,, which was made immediately after
the final deviation in the activity specification or level of effort occurred.
The Z values computed for a particular estimator can then be tabulated and
studied, as additional activities with which he is associated are completed. Theo-
retically, these Z values should vary randomly about zero from about - 3 to +3,
with the majority of the values near zero. Deviations from this pattern have a
very logical interpretation; they should be studied by the estimator who supplied
the data so that improvements in the estimates can be made. If the points vary
about a mean less than the zero centerline, then the estimator has the common
problem of underestimating the time required to perform assigned tasks. Simi-
larly, variation about a mean greater than zero would indicate a consistent over-
estimation of performance times. Also, the points should vary from - 3 to +3,
or possibly - 4 to + 4 if the performance time distributions are skewed. If they
vary over too wide a range, say ? 5 , then generally, the difference between the
pessimistic and optimistic estimates (b - a) must be increased. The other possi-
bility is not too likely, but consistent variation over too narrow a range would
call for a reduction in the difference (b - a).
 The PERT Statistical Approach 293

SUMMARY OF HAND PROBABILITY COMPUTATIONS

PERT probability computations can be handled in much the same way as time-
cost trade-offs were handled in Chapter 8. If a computer is being used, then
estimates of a, m , and b must be obtained and used on all activities. In this
case, probabilities of meeting specified scheduled times are given in the com-
puter output, such as are shown in Figure 9-13, for the network shown in
Figure 9-12. However, if the computations are being made by hand, one could
start off with a single time estimate for the mean performance time for each
activity, and then obtain a, m , and b values only for those activities on the criti-
cal path, or the longest path leading to the scheduled event in question. This
procedure is summarized below.

1. Make the usual forward and backward pass computations based on a single-
time estimate, D, for each activity.
2. Suppose one wishes to compute the probability of meeting a specified
scheduled time for event X. Then obtain estimates of a, m and b for only
those activities that comprise the "longest path" from the initial event to
the event X. If necessary, adjust the length of this path as dictated by the
new t , values based on a, m and b.
3. Compute the variance for event X, (VT), by summing the variances for the
activities listed in Step 2. VT = sum of values of [(b- a)/3.2I2,for each
activity on the "longest path" leading to event X.
4. Compute z using equation (5) and look up the corresponding probability
in the normal curve table given in Appendix 9-1.

To use the PNET method described below, values of a and b must also be col-
lected on the "near" critical paths, to be defined.
Referring again to Figure 9-12, the practice of labeling the events on the net-
work, rather than the activities, should be noted. Event labeling in this manner
is not recommended because it confuses the reader; one cannot be sure which
arrows and time estimates apply to which events. This procedure was adopted
to be compatible with the early government practice of controlling this type of
project by monitoring milestone events. Milestones are, of course, key points
in time in the life of a project and are, therefore, events which occur at the com-
pletion of one or more activities. Upper management keys on the status of mile-
stone events, while lower levels of management concentrate on the performance
of individual activities as shown in Figure 9-13. To facilitate this reporting at
several levels of detail, computer programs are utilized to summarize networks
by retaining only the specified set of milestone events. (This concept was dis-
cussed under Network Condensation in Chapter 3.)
294 11 I Advanced Topics
 PAGE 1
PERT SYSTEM
PERT SYSTEM

RUN 1 ENDING EVENl

BY PATHS OF CRITICALITY DATE 06-07-61
CHART AJ LR SN 9 ELECTRONIC MODULE (ILLUSTRATIVE NETWORK) SYSTEM W 034
EVENT DATE DATE EXP EXP
PREDECESSOR SUCCESSOR NOMENCLATURE DEP EXPECTED ALLOWED SCHD/ACT PROB SLACK TIME VAR

4004-715 4004-716 REV DATE (SMITH TROMBONE RECO-BEG TEST) 98 05-26-61 A06-07-61 -1.6 t
4004-716 4004-714 SMITH TROMBONE TESTED 0146 06-23-61 06-12-61 -1.6 + 2.3 .4
4004-714 4004-727 TROMBONE CHOSEN-BEGIN MECH OESIGN 0146 06-28-61 06-16-61 -1.6 t 3.0 .5
4004-727 4004-728 RFP PARAMP FLIGHT HARDWARE 0146 07-20-61 07-08-61 - 1.6 + 6.1 .7
4004.728 4004-729 PARAMP CONTRACT LET 0146 08-17-61 08-05-61 -1.6 t10.1 .8
4004-729 4004-730 PARAMP RECEIVED 10-26-61 10-14-61 -1.6 t20.1 1.3
4004-730 4004-734 PARAMP TESTED 0146 11-09-61 10-28-61 -1.6 t22.1 1.4
4004-734 4004-742 CHOICE BETWEEN PREAPIP-PARAMP 0146 11-13-61 11-01-61 -1.6 t22.6 1.4
4004-742 4004-743 COMPL MICROWAVE ASSY 0146 12-09-61 11-28-61 -1.6 t26.5 1.6
4004-743 4004-199 COMPL FINAL TEST MICWAVE ASSY-DELIVERED 0146 12-25-61 12-13-61 12-13-61 .12 -1.6 t28.6 1.9

4000-001 4004-713 REV DATE (JONES TROMBONE RECO-BEG TEST) 99 05-29-61 A06-07-61 -1.3 +
z
4004-713

4004-714
4004-714 JONES TROMBONE TESTED

4004-717 TROMBONE CHOSEN-BEGIN MICWAVE ASSY TEST

0146 06-21-61 06-12-61

0146 06-28-61 07-02-61

-1.3

t .5
t

t
2.0

3.0
.1
.5
m

p
-
rn
4004-717 4004-731 RFP FOR FLIGHT HDW-MIXER-TROMB-PREAMP 0146 07-20-61 07-24-61 + .5 t 6.1 .7 4
4004-717 4004-739 COMPL MICWAVE ASSY TEST-RFP LOC OSCIL 0146 07-20-61 07-24-61 t .5 + 6.1 .7 J?
4004-731 4004-735 TROMBONE CONTRACT LET 0146 08-17-61 08-21-61 t .5 t10.1 .8 g,
4004-731 4004-737 MIXER CONTRACT LET 0146 08-17- 61 08-21-61 + .5 +10.1 .B a
4004-739 4004-740 CONTRACT LET FOR LOC OSCIL AND PWR SPLT 0146 08-17-61 08-21- 61 t .5 t10.1 .8 "
4004-735 4004-736 TROMBONE RECEIVED 10-14-61 10-18-61 + .5 t18.5 1.8
4004-737 4004-738 MIXER RECEIVED 10-14-61 10-18-61 + .5 +1B.5 1.8 u
4004-740
4004-736
4004-741
4004-742
LOC OSC-PWR SPLITTER RECEIVED
TROMBONE TESTED
10-14-61
0146 10-28-61
10-18-61
11-01-61
+ .5
+ .5
t18.5
t20.5
1.8
1.9
5g
4004-738 4004-742 MIXER TESTED 0146 10-28-61 11-01-61 + .5 t20.5 1:9 s
4004-741 4004-742 LOC OSC-PWR SPLITTER TESTED 0146 10-28-61 11-01-6: + .5 i20.5 1.9
Figure 9-13 Typical PERT computer output. (First three-paths shown here.)
I
m
296 11 I Advanced Topics

The critical path through the project network in Figure 9-12 is marked by the
heavy line. The computer output in Figure 9-13 is based on a primary sort of
slack, and a secondary sort of early start time. This places the critical path as
the first set of 10 activities, the next most critical path as the next set of 2 ac-
tivities, and the third most critical paths as the final set of 12 activities. Notice
that the slack on the critical path is - 1.6 weeks, because the scheduled comple-
tion time (12-1361) is earlier than the expected time (12-25-61). Because of
this, the probabiliy of meeting this schedule is a low value of 0.12. Clearly, this
would be a red Iight signal, calling for some immediate application of time-cost
trade-off procedures to raise this probability to an acceptable level. Further-
more, it would be unwise to wait further t o consider this problem, because as
time goes by, the most economical way t o buy time may have passed by.

INVESTIGATION OF THE MERGE EVENT BIAS PROBLEM

As pointed out in the introduction to this chapter, the conventional PERT pro-
cedure described above always leads to an optimistically biased estimate of the
earliest (expected) occurrence time for the network events. This bias arises be-
cause all subcritical paths are ignored in making the forward pass computations.
If the longer path leading to a merge event is much longer than the second
longest path, and/or the variance of the activities on the longest path is small,
this bias will be insignificant. The first part of this section will be devoted to a
series of examples designed to give the reader an appreciation for the signifi-
cance of this bias problem. Finally, a solution t o this problem using the PNET
procedure and Monte Carlo simulation will be described in some detail.

Magnitude of Bias

A study of this problem was made by MacCrimmon and Ryavec,' who con-
sidered two of the more important factors affecting the magnitude of the merge
event bias. First, one would intuitively expect the bias t o increase as the num-
ber of parallel paths to the network end event increases. This is studied in Fig-
ure 9-14 below. Second, one would also expect the bias to increase as the ex-
pected length of the parallel paths become equal. This is studied in Figure 9-15.
Consider the four activity network in Figure 9-14a. The particular discrete
distribution assumed for each of these activities can be identified in Table 9-3 by
the corresponding mean shown on the network activities. There are two paths,
ABD and ACD, both having a mean length of 6. The mean of the maximum
time distribution, or the earliest expected time for event D is 6.89. Thus, the
error in the PERT calculated mean is 12.9 percent of the actual mean.
There are two possible ways a third path, with a mean length of 6, may be
created by adding one more activity. In one case the path may be completely
independent of the other two paths, thus resulting in a third parallel element,
 The PERT Statistical Approach 297

PERT (mean1 6.00 6.00 6.00

Exact (mean1 6.89 7.34 7.07

Error .12.99h -18.2% -15.256

Figure 9-14 Effect of parallel paths. with and without correlation,on themergeevent bias.

AD, as depicted in Figure 9-14b. Alternatively, an activity BC can be added

with a mean time of 2, thus creating path ABCD, shown in Figure 9-14c. In
both cases there are three paths, all of mean length 6, and the network has four
events and five activities.
The addition of the third path in parallel leads to an increase in the deviation
of the PERT-calculated mean (still 6) from the actual mean, which in this case
is 7.336. Thus, the error has increased to 18.2 percent. Figure 9-14c, on the
other hand, is a network configuration, where there is a cross connection be-
tween two parallel paths. Since there are three paths, one would expect a larger
error than in a similar network with only two paths (such as Figure 9-14a), al-
though not as large an error as in Figure 9-14b, where the three paths are in
parallel. The correlation (resulting from the common activities) in the network
of Figure 9-14c does indeed have the effect discussed, and the mean of the maxi-
mum time distribution lies between these two bounds, being 7.074. The error
as a percent of the actual mean is 15.2 percent.
The examples given above are extreme cases, since all the paths have the same
expected duration-hence, they are all critical paths. If the durations of some
paths are shorter than the duration of the longest path, their effect on the proj-
ect mean and standard deviation would not be as great. However, if they have a
mean duration very close to the mean duration of the critical path, they would
not be critical but they would have ;m effect almost as significant as the exarn-
ples of the previous sections. The following examples shown in Figure 9-15 indi-
cate the effect of slack in a path length.
The simple network has only two paths, ABC and AC. AU activities are as-
sumed to be normally distributed, with standard deviation equal to 1, and the
appropriate mean given on the diagram. It may be noted from the diagrams that
various lengths are assumed for paths ABC and AC, ranging from both of them
being of equal length, to path AC being only $ the length of path ABC.
 Id)

Ratio of path AC
Lengths: path ABC
PERT (Mean)
Exact (Mean)
% Error
PERT (Std. Dev.)
Exact (Std. Dev.)
% Error
Figure 9-15 Effect of slack on merge event bias.
 The PERT Statistical Approach 299

Table 93. Discrete Distribution for Activities in Figure 9-14

t Probability t Probability

1 % 2 x
2 K 4 %
3 % 6 %
Mean = te = 2 Mean = re =4
S t d . Dev. = ( v ~ =)0.707
~ ~ ~ S t d . Dev. = = 1.414
Coef. of Var. = ( ~ ~ ) =~35%
/ ~ l t ~ Coef. of Var. = ( ~ ~ l ' =/ 35%
~ / t ~

This example indicates that the deviation of the PERT-calculated mean and
standard deviation, from the actual mean and standard deviation, may be quite
large when the paths are about equal in length, but the difference decreases
substantially as the path lengths become farther apart.

Rules of Thumb on Merge Point Bias

To summarize qualitatively the above results on merge event bias, it can be

noted that t h e event increases-
as

1. the number of merging activity increases,

2. the expected complete times of the merging activities get closer together,
3. the variances of the merging activities increase, and
4. the correlation among the merging activity complete times approacheszero.

Because of point 2 above, the correction at most merge events will be negligible
and thus can be ignored. From a study of tables derived by C l a ~ k giving
,~ the
expected value of the greatest of a finite set of random variables, this can be
stated as a useful rule of thumb as follows.

Rule:

If the difference between the expected complete times of the two merging
a$yitie-considered is greater than the larger of their respective standard
deviations, then the bias correction will -.
be small;
-if the difference is greater
than two standard deviations, the bias will be less th& a few percent and can
be ignored. (The difference referred to here is what has been defined in Chapter
4 as activity free slack.) If there are more than two merging activities, this rule
should be applied to the two with the latest expected finish times.
The validity of this rule is illustrated in Figure 9-15. The difference (slack)
in the expected time of the two merging activities is less than one standard devia-
300 11 I Advanced Topicr

tion in Figure 9-15b and is greater than two in Figure 9-15c. The corresponding
biases of 8 percent and 0.5 percent are appreciable andinsignificant,respectively,
as suggested by this rule.
If the above does not rule out the need for a bias correction, then it should be
made by one of the procedures described below.

Analytical Merge Event Bias Correction Procedures

The merge event bias correction problem is essentially a statistical problem, deal-
ing with a random variable defined as the maximum value of a set of random
variables, not necessarily statistically independent. The latter condition compli-
cates the problem greatly. The maximum value is the earliest expected occur-
rence time of the (merge) event in question, and the set of random variables is
the actual complete times of the activities merging to the event in question.
These latter times are not always statistically independent, because of the net-
work crossover condition previously illustrated in Figure 9-14c.
This is an intriguing statistical problem that has caught the fancy of many
researchers, and dozens of papers have been written about it. Only the rela-
tively recent work of Ang, Abdelnour and Chaker' (1974) will be presented
here because it is the simplest of the procedures that produce a satisfactory
solution. Their procedure is called PNET.
To introduce this procedure, consider again the network shown earlier in
Figure 9-1. The conventional critical path is 0-3-7-8, having an expected time
of 15, with a standard deviation 1.68 days. Suppose the three time estimates for
the additional activities 3-4, 4-5, and 5-8 are as shown in Table 9-4. These data

Table 9-4. Mean and Standard Deviation of the Critical and Near Critical Paths
for the Network in Figure 9-1

Time Estimates Path 0-3-7-8 Path 0-3-4-5-8

Activity a m b Mean Variance Mean Variance

0-3 1 2 3
3-7 6 8 10
7-8 3.5 5 6.5
3-4 1 4 13
4-5 2 4 6
5-8 2 3 4
~otals*
Standard Deviation

'The mean and variance of the duration of a path is merely the sum of the means and vari-
ances of the activities along the path in question; the standard deviation of the path dura-
tion is then obtained as the square root of its variance.
 The PERT Statistical Approach 301

lead to the mean, variance, and standard deviation statistics for the critical path,
0-3-7-8, and the near critical path, 0-3-4-5-8, as shown at the bottom of the
table. Using eq. (S), the cumulative probability curves for each of these paths
are then computed and are shown in Figure 9-16. They appear as straight limes
because the graph paper used here, called Normal Probability paper, has a verti-
cal scale adjusted to produce a straight line for any normally distributed random
variable. It is obvious from these two graphs that the PERT probability proce-
dure, which considers only path 0-3-7-8, is biased on the high side. In fact, for
any scheduled time greater than the cross over point (about 15.7) the cumulative
probability for the near critical path is considerably less than for the critical path
itself. This anomoly occurs because of the very large variance for the near criti-
cal path; a situation which could occur in practice. It should also be noted that
although the individual path lengths have Normal distributions (by virtue of the
Central Limit Theorem) the distribution of the project duration, considering all
paths, is not Normal. This distribution, t o be derived below, is shown by the
dashed line in Figure 9-16. This censoring of the lower tail of the distribution
is a characteristic result which has been verified repeatedly by simulation. It can
be explained by recalling that the actual project duration is the maximum of a
set of random variables. The lower tail of this distribution of the maximums is

.999 -
,998 -
,995 -
.99
-
-
.98
-- -
--
.95
--
.90

--
.80
> .E -

Standard

1 1 l 1 1 1 1 1
6 8 10 12 14 16 18 20 22 24 26
Project scheduled duration time. Ts
Figure 916 Cumulative probability of path and project mmpletion on o r before time T5.
302 11 I Advanced Topics

truncated because a low maximum requires all of the paths t o be low, and hence
it occurs infrequently. However, a high maximum will result when only one or
more high path times occur, and hence it occurs more frequently. At this point
it is appropriate to ask, "What is the correct cumulative probability?"
If we are jistified in ignoring all but these two paths, we could proceed t o
answer this question by first assuming they are statistically independent. This is
a very good assumption because these paths have only one element in common,
activity 0-3, and its variance is negligible in comparison to the total path vari-
ances. Thus, from basid probability theory of independent events, the prob-
ability that the project will be completed on or before time T, is merely the
product of the cumulative probabilities that each path wiU be completed on or
before time T,. For example, for T, = 14, this probability (reading from Figure
9-16) is 0.50 X 0.27 = 0,135. The latter value is shown by the dashed curve, la-
beled Both Paths. This is essentially the procedure developed by Ang,' called
PNET. It amounts to fmding the set of assumed independent paths which "rep-
resent" the network, and then set the cumulative probability for the project
equal to the product of the cumulative probabilities for each of the "represen-
tative" paths. It can be shown that the dashed PNET curve in Figure 9-16 is
almost identical t o the correct curve.

ANG'S PNET ALGORITHM

The PNET algorithm utilizes the important fact that the linear correlation coeffi-
cient between any pair of path lengths, say Tt and Ti for paths i and j, is given
by

where the sum in the numerator is the sum of the variances of the activities that
are common t o both paths i andj, and the denominator is merely the product
'of the standard deviations of the total durations of paths i and j. The value of
pij varies from 0 for completely independent paths, to 1 for identical paths. For
the two paths in the above example, the data in Table 9-4 shows that pii is only
0.057, indicating almost complete independence of these two paths.

In the PNET algorithm, this correlation coefficient is used t o determine the set
of representative paths for the network. Jftwo u a w e a correlation coeffi-
w e r than 0.5. then the oln-_fhtew
t as is said t o represent both
paths. and the shorter oath is dropped from consideration. Similarly, if the cor-
relation coefficient is less than 0.5, the two paths are assumed in@pzd_entmd
both remain under consideration. Then, the probability that a project will meet
 The PERT Statistical Approach 303

Mean = t,
Variance
I

Figure 9-17 lllustrativc network for Monte Carlo simulation and PNET.

a specified completion date is taken as the product of the probabilities that each
of the representative paths will meet the specified completion date.
An application of this procedure, given in the next section, shows excellent
agreement with the correct results obtained by simulation. Ang gives three ap-
plications of this procedure, and each shows excellent agreement with the results
obtained by simulation. One explanation of why this procedure seems t o work
better than expected, based on intuition, is that a form of the Central Limit
Theorem is in operation. That is, dropping paths with correlations greater than
0.5, tends to cancel the error effects of assuming paths with correlations less
than 0.5 are independent. A modification of Ang's - original
- PNET procedure is
given below, along with its application to the network shown above in Figure
9-1 7.

Modified PNET Algorithm

1. Generate the list of major network paths* sequentially with decreasing

mean path durations, starting with the critical path(s). In case of ties, arranRe
the paths with decreasing standard deviations. The major paths are defmed as
those whose mean path durations are at least a certain percentage of the ex-
pected critical path duration. (The rule developed in the previous section and
adopted here,. suggests
-- the retention of all paths whose lengths differ from
the critical path by less than twice the larger of the standard deviations of the
two paths.) Number the major paths from I' to N. This step is illustrated for
the network shown in Figure 9-17, and the results are given in Table 9-5. All

*The listing of paths in order of criticality could be accomplished from a slack sort of the
project activities, moditied so that each slack path extends from the initial to the terminal
network event.
304 II I Advanced Topics

Table 9-5. Application of PNET to the Network in

Figure 9-17.
-- -- -- -

Standard
No. Path Meen Variance Deviation

10 network paths are shown here for illustrative purposes; the listing could
have stopped after generating path 6 and showing that its mean differs from
the first (critical) path by 66 - 50.3 = 15.7, which exceeds twice the larger of
the path standard deviations, i.e., 2 X 7.8 = 15.6.
2. Compute the correlation coefficient for each pair of major paths using
equation (7), and arrange them in matrix form. For the above example, this
matrix is shown in Table 9-6.
3. Determine the set of representative paths for the network. This proce-
dure is given for the correlation matrix shown in Table 9-6.

Table 9-6. Correlation Matrix for the Five Major Paths in the Network
in Fiaure 9-17
1 . . . . . . .. . . . . . 3 . . . ... . . . . . . . . . tRepresentative Set of Paths
\ i .
i
I
& , 4
I
I
I5 t M a j o r Paths Considered

- -
Paths Deleted
 The PERT Statistical Appmach 305

3.1. Start with path 1, enter it into the set of representative paths, and
move across row 1 of the correlation matrix deleting all paths with correlations
greater than 0.5. In this example, delete paths 2 and 5 because r,, = 0.88 and
rl, = 0.86 are both greater than 0.5.
3.2. Move t o the next remaining path, enter it into the set of representative
paths, and move across this row of the correlation matrix again deleting all paths
with correlations greater than 0.5. In this example, enter path 3 into the set of
representative paths, and delete path 4 because r , = 0.71 is greater than 0.5.
3.3. Continue Step 3.2 until there are no remaining major paths to be con-
sidered; at this point stop, list the set of representative paths and number them
.
1 , . , K. In this example the procedure stops after Step 3.2 because no
major paths remain to be considered, and the set of representative paths con-
sists of paths 1 and 3.
4. Calculate the probability that the duration of each representative path
(TI, . . . , TK) will not exceed some scheduled project duration (T,), assuming
that each path duration has a normal distribution with mean and standard devia-
tion computed in Step 1, i.e., calculate P(T, 5 T,), . . . ,P(TK 5 T,). These
probabilities are shown on the left side of Table 9-7, for the example problem.
5. The probability that the project completion time, T, will meet the sched-
ule time, T,, is denoted by P(T 5 T,), and is then given approximately by the
product

P(T 5 T,) P(Tl 5 T,) X . . . X P(TK 5 T,).

For the example, there are two representative paths, and this probability is given
as follows:
PIT 5 T,) =P{Z 5 (T, - 66)/7.8} X P{Z 5 (T, - 58)/4.1}
The results of these computations are given in Table 9-7. I t will be shown
below, that these PNET probability estimates are not significantly different from
the results obtained by Monte Carlo simulation, where the latter are taken as the
"correct" probabilities. The computation of the estimates of the PNET mean
and standard deviation are also shown at the bottom of Table 9-7.

Monte Carlo Simulation Approach to Pert Probabilities

The Monte Carlo simulation approach to the solution of this problem was used
by Van Slykel2 in 1963. He recognized this problem as one of simply solving
the stochastic network model to find something that corresponds, in some sense,
to the project duration and critical path in the deterministic case. The difficulty
here was avoided to some extent by approximating the random problem by a
series of problems of the deterministic form. To accomplish this, Monte Carlo
simulation was used. A bonus from this approach was that it not only gave
306 11I Advanced Topics

Table 9-7. Computation of PNET Project Probabilities from Representative

Path Probabilities

Path 1 = 1-2-4-5-9 Path 3 = 1-3-5-9

z1 = 23 = PNET =
Ts IT- 66117.8 P ( ~ d . I7 ~ (T, - 58)14.1 P(Z 5 rs) PV 521) . P ( z ~ z ~ )

MEAN 66 58 66.9'

STANDARD
DEVIATION 7.8

his mean value was computed from the PNET cumulative probability distribution using
the following formula:
K
Mean = t; X p (ti)
i=l

.
where ti is the class mark (mid-point) of the i t h class interval ( i = 1.2.. . , K; where K = 7
in this sxample), and p ( t i ) is the probability that the (random) project duration will fall in
the i t h class interval.

-
Mean = LO18 - 0) 52.5 + (.I52 - ,018) 57.5 + .
+ (1.OM) - .9931 87.5 66.9
- - + (.993 - ,963) 82.5

-
The standard deviation, or variance, was computed in a similar fashion, replacing the class
mark (e.g., 52.51, by the (class markmean)2 (52.5 - 86.9)'.

unbiased estimates of the mean and variance of the project duration, along with
the distribution of total project time, but it also gave estimates for quantities
not obtainable from the standard PERT approach. In particular, the 'criticality'
of an activity, i.e., the probability of an activity being on the critical path, can
be calculated. One of the more misleading aspects of conventional PERT meth-
ods is the implication that there is a unique critical path. In general, any of
a number of paths could be critical, depending on the particular realization of
the random activity durations that actually occur. Thus, it makes sense to talk
about a criticality index. This appears to be an exceedingly useful measure of
the degree of attention on activity should receive by management, and is not as
misleading as the critical path concept used in PERT. It should be added that
the probability of an activity being on the critical path is not correlated too well
with slack, as computed by the conventional PERT procedure, which is the fac-
 The PERT Statistical Approach 307

tor that usually determines the degree of attention that a particular activity
receives.
The Monte Carlo simulation procedure was applied by Van Slyke to the net-
work given in Figure 9-17. Each activity was assumed t o have a beta distribution
with mean, t,, and a variance, V,, as noted on each activity. As generally recom-
mended by Van Slyke for this purpose, 10,000 sets of random times were gen-
erated for each activity in the network. For each of these sets, the longest path
through the network was determined: its duration was noted, as well as a count
for each activity on the critical path. The results of these 10,000 simulations are
given in Figure 9-18, where the probability that an activity was on the critical
path is noted on each activity. For example, 0.737 on activity 1-2 means that in
7370 of the 10,000 simulations, this activity was on the longest path in the net-
work. Also, given at the bottom of Figure 9-18 are the statistics pertaining t o
the total project duration. We note here that the PERT estimate of the project
mean was low (optimistic) by only 1.5 percent; however, the variance was esti-
mated too high by 42 percent, and the results previously obtained by PNET are
well inside the 95% confidence intervals based on the simulation estimates. The
PERT result is about as expected according to the rule @en in the previous sec-
tion, because there is a considerable amount of slack along the subcritical
path(s) at each merge event. It is interesting to note, however, that activities 1-3
and 3-5, which are not on the conventional PERT critical path, have appreciable
probability of ending up on the actual critical path.
Another output of the Monte Carlo simulation study is given in Figure 9-19,
where the cumulative probability of a specified project duration is given for sim-
ulation and PNET, which were almost identical, and is compared with the results
given by the conventional PERT procedure.

SIMULATION SAMPLE SIZE

A final note on the simulation procedure regards the cost of obtaining these
results. A sample size of 10,000 simulations was used in the example in Figures
9-17 and 9-18, however, this is considerably larger than justified for this problem.
A recent empirical study by Crandalls indicates that this sample size can be re-
duced to 1000 or less and still obtain an adequate level of confidence in the final
estimates. Other practitioners of simulation suggest samples as low as 400. This
is the generally recommended range of sample sizes for this problem.
A statistical analysis for the criticality index can be based on the binomial dis-
tribution, since for each simulation trial, any particular activity is either on, or
not on, the critical path. The (1-a) level confidence interval could then be con-
structed on the true criticality index, p, from the equation
308 11 I Advanced Topics
 The PERT Statistical Approach 309

1.0
-I - ,

0.8 -

0.6 -

0.4-
---PERT
Monte Carlo
and PNET

I
50 55 60 65 70 75 80 85 90

Project duration

Figure 9-19 Cumulative probability of project duration.

where p^ is the estimated criticality index, n = sample size, and Z,/, is the stan-
dard normal deviate. Application of this formula at the 95% confidence level
(Z,,,,, = 1.96) would indicate that if p is in the 10% to 90% range, and sample
sizes in the 400 t o 1000 range, then the estimates would fall in the ranges
*
(100 p + 21% to (100 p 5)%. Estimates within 2% t o 5% certainly represent
an acceptable level of accuracy.
Regarding the estimation of the mean project duration, some assumption
must be made regarding the coefficient of variation (CV) of the project dura-
tion, where CV = (Standard DeviationIMean) X 100%. From the real examples
reported by Ang' , and many others reported in the literature, the CV was found
to vary from 5% t o 15%. In this case, the width of the 95% confidence interval
on the estimate of the mean will vary from f0.5 to 21.5% of the mean for sam-
ple size 400, and f0.3 to 51.0% for sample size 1000. This can be computed
from the formula:

The estimation of the corresponding standard deviation of project duration

would not be quite as good, ranging from f 4 to f7% of the true value. Simi-
larly, this can be computed from the following formula, which again indicates
that the error levels are quite acceptable.

Finally, a confidence interval could be placed on the cumulative probability

curve by the use of the Kolmogorov-smirnov6 test. Essentially, this method
states that the observed cumulative distribution curve will not deviate from
the true curve by more than 1.22/@, 1 . 3 6 / n , or 1.63/fl, at the 90%, 95%,
310 11 I Advanced Topics

and 99% confidence levels, respectively. For example, a confidence interval

statement can be made about the observed cumulative distribution curve (solid
line) shown in Figure 9-19, which was based on a sample size of N = 10,000.
That is, we can state with 95% confidence, that the entire observed cumulative
distribution does not deviate from the true curve of more than 1.361-=
0.0136, or about 1.4%. If a sample size of only 400 or 200 had been used, this
maximum deviation would increase t o 6.8% and 9.6%, respectively. The latter
is about as large as one could consider tolerable and thus N = 200 represents
about the smallest sample size that should be used.
With these recommended sample sizes of 400 to 1000, a number of simula-
tion computer programs are available that could handle most networks at an
affordable cost. Where cost is a severe limitation, a sample size as small as 200
could even be used, however, values below this would not give reliable estimates
of the shape of the distribution of project durations. The use of GERT 111 for
this type of simulation will be illustrated in Chapter 10, Figures 10-4 through
10-7.
The application of PNET to precedence diagramming is treated in exercise 11.

SUMMARY

In this chapter, the PERT statistical approach t o project planning and control
was given, which leads to a probability that a given scheduled event occurrence
time will be met, without having to expedite the project. The conventional
PERT procedure derives its measure of uncertainty in the event occurrence times
from the three performance time estimates, optimistic, pessimistic, and most
likely, for each network activity. This procedure was modified by defming the
optimistic and pessimistic times as 5 and 95 percentiles, respectively, of the
hypothetical activity performance time distribution, rather than the end points
of the distribution. Based on the Central Limit Theorem, estimates of the mean
and variance in activity performance times were then used to compute a prob-
ability of meeting arbitrary scheduled times for special network events. It was
recognized that it is difficult to obtain accurate estimates of the activity perfor-
mance times, and procedures for improving the estimation by feedback of past
estimation performance plotted on a standardized control chart was outlined.
The merge event bias, introduced in conventional PERT by ignoring all but
the critical path, was then discussed with examples illustrating the magnitude of
this problem. A simple rule of thumb was given to determine, from the com-
pleted conventional PERT analysis, whether this bias will be serious or not.
This rule simply states that at merge events, one can ignore the path with free
slack in excess of twice the larger of the two paths standard deviations.
Two practical procedures to overcome the statistical errors in conventional
PERT were described, that is, PNET and Monte Carlo simulation. Both methods
 The PERT Statistical Approach 311

will give accurate estimates of the expected project duration and the probability
of meeting a range of scheduled duration times. It is, however, easier t o collect
this information on intermediate network events with simulation than with
PNET. Also, simulation alone will give the criticality index on each project
activity. Something approaching this might be achieved with PNET if it were
programmed to give the cumulative probability curves for the critical and near
critical paths, but this would not be as easy t o interpret as the criticality index.
Finally, it has been estimated by Angl that the computer cost of simulation is
about an order of magnitude greater than running PNET.
From the above comparison, it appears that PNET should become a standard
procedure, expected of all PERT computer routines. If the importance of the
project warrants the additional expenditure, then simulation should be utilized,
primarily t o give information on all milestone network events, and the criticality
index on all activities. With these improvements in PERT methodology in plade,
interest in this technique just may be rekindled.

REFERENCES

1. Ang, A. H-S, J. Abdelnour, and A. A. Chaker, "Analysis of Activity Net-

works Under Uncertainty," Journal o f the Engineering Mechanics Division,
Proceedings of American Society of Civil Engineers, Vol. 101, No. EM4,
August 1975, pp. 373-387.
2. Clark, C. E., "The PERT Model for the Distribution of an Activity Time,"
Operations Research, Vol. 10, No. 3, May-June 1962, pp. 405 and 406.
3. Clark, C. E., "The Greatest of a Finite Set of Random Variables," Opera-
tions Research, Vol. 9, No. 2, March-April 1961, pp. 145-162.
4. Crandall, K. C., "Probabilistic Time Scheduling," Journal o f the Constnrc-
tion Division, Proceedings of American Society o f Civil Engineers, Vol. 102,
No. C03, Sept. 1976, pp. 415-423.
5. Crandall, K. C., "Analysis of Schedule Simulations," Journal o f the Con-
struction Division, Proceedings of American Society o f Civil Engineers, Vol.
103, No. C03, Sept. 1977, pp. 387-394.
6. Kendall, M. G. and A. S. Stuart, "The Advanced Theory of Statistics,"
Griffin, London, Vol. 2, 1973, p. 473.
7. King, W. R. and T. A. Wilson, "Subjective Time Estimates in Critical Path
Planning-A Preliminary Analysis," Management Science, Vol. 13, No. 5,
January 1967, pp. 307-320.
8. MacCrimmon, K. R. and C. A. Ryavec, "An Analytical Study of the PERT
Assumptions," Operations Research, Vol. 12, No. 1, January-February
1964, pp. 16-37.
9. Malcolm, D. G., J. H. Roseboom, C. E. Clark, and W. Faza, "Applications
of a Technique for R and D Program Evaluation," (PERT) Operations
Research, Vol. VII, No. 5, September-October 1959, pp. 646-669.
312 11 I Advanced Topics

10. Moder, J. J. and E. G. Rodgers, "Judgment Estimates of the Moments of

PERT Type Distributions," Management Science, Vol. 15, No. 2, October
1968, pp. B-76, B-83.
11. PERT, Program Evaluation Research Task, Phase I Summary Report, Special
Projects Office, Bureau of Ordnance, Department of the Navy, Washington,
July, 1958.
12. Van Slyke, R. M., "Monte Carlo Methods and the PERT Problem," Opera-
tions Research, Vol. 11, No. 5, September-October 1963, pp. 839-860.

EXERCISES

1. Verify the Central Limit Theorem by sampling, that is, by tossing three dice
and recording the results on graphs, such as those shown in Figure 9-6. For
convenience, use a white, a red, and a green die. Call the number of spots
on the white die, X, and plot on the first figme. Call the number of spots
on the white plus the red die, Y, and plot on the second figure. Finally, call
the number of spots on all three dice, Z, and plot on the thud figure. Com-
pare the results of your experiment with the theoretical values given in
Figure 9-6.
2. Verify the expected time and variance for event (4004-1 99), and verify the
the probability of 0.12 given for activity (4004-743)-(4004-199) in Figure
9-13b. Note: the values of a and b given in this figure are the end points of
the distribution of activity performance time as used in conventional PERT
Hence (vt)ll' = ( b - a)/6 should be used in place of equation (3) given in
this text. Also, note that the time interval 12-13-61 to 12-25-61 is equiva-
lent to 1.6 working weeks.
3. Consider the oversimplified network given in Figure 9-20 which might be
only a portion of a larger network, a portion which is subject t o consider-
able chance variation in the performance times. In Figure 9-20 a and b are
5 and 95 percentiles, respectively.
a. Compute t , and V , for each of the four activities.
b. What is the earliest expected time of event 3?

Figure 9-20
 The PERT Statistical Approach 313

c. What is the variance, VT,for the actual occurrence time for event 3?
d. What is the probability that the project will be completed by time 8?
By time 9?
e. What time are you fairly sure (say 95 percent confident) of meeting for
the completion of the project, i.e., the occurrence of event 3?
4. With reference t o Figure 9-1, what is the probability that event 8 will be
completed on or before the end of the 12th day, assuming all activities are
started as early as possible? What scheduled time for the entire project
would you feel 95 percent confident of meeting without having t o expedite
the project?
5. With reference t o Figure 9-1, what is the (conditional) probability that event
8 will be completed on or before the end of the 16th day, assuming that a
scheduled time of 10 is met on event 7?
6. Repeat exercise 5 on the assumption that event 7 has occurred 2 days late,
i.e., at the end of the 12th day.
7. Solve exercise 3 using the Monte Carlo approach described in the text. For
simplicity, assume that the distribution of perfoqnance times for each ac-
tivity is rectangular between the limits given by a and b . Using a table of
random numbers that are uniformly distributed on the range from 0 to 1,
given in most standard statistics texts, transform them t o the distribution re-
quired for each of the activities in Figure 9-20. For example, activity 2-3
has a possible range of 3 t o 7, or 4 time units, and a mean of 5. Letting r
denote a random number uniformly distributed on the interval 0 t o 1, then,
t = 3 + 4 r , will be uniformly distributed on the range 3 t o 7 as desired. De-
cide for yourself how many simulations of the network should be made,
and express the results in a form such as given in Figures 9-18 and 9-19.
Discuss your results.
8. Repeat exercise 7 for the network given in Figure 9-1 making the same as-
sumptions as suggested in exercise 7 about the distributions of activity per-
formance times. To carry out this exercise, use the following values of a, m ,
and b.

Activity a m b
 a. Applying the rule of thumb given in the text on merge event bias, will
the latter be significant in this problem? Why? At what merge events?
b. If we wanted to simplify the Monte Carlo simulation, would the elimina-
tion of slack paths 0-6-7 and 0-1-2-5 be justified?
c. Perform the Monte Carlo simulation and comment on the results obtained.
9. Referring to exercise 2 in Chapter 8 assume the times given under a and b
are 0 and 100 percentiles, so that the standard deviation, (vt)'I2 = (b - a)/6,
and the times under the m column are single time estimates of the mean,
i.e., t , = m , in this case. Without making a complete PERT statistical anal-
ysis, answer the following questions.
a. Is there a significant merge event bias problem for this network?
b. What is the mean and approximate standard deviation of the total time
for this project?
c. What are the approxitt~atechances of completing this project in 400
hours?
10. Apply the PNET procedure to the network shown in Figures 9-12 and 9-13.
a. Compute the corrected probability of meeting the schedule of 12/13/61
for event 4199. (This is equivalent t o a project duration of 27.0 weeks).
b. Compute the cumulative probability curve for a range of scheduled times
from 26 to 36 weeks.
c. What project scheduled duration currently has a probability of 0.90 of
being met?
11. How would PERT be applied t o precedence diagramming? Answer this
question by applying the Central Limit Theorem.
a. Describe a procedure that could be used assuming the critical path is
specified as suggested in Chapter 4. For example, the network in Figure
4-1 If has the following critical path. Framing(NC)-FFl-Electrical(RC)-
SS1-Finishing(NC). Assuming estimates of the mean and variance for
each activity duration are available, give equations for the mean and vari-
ance for the project duration, T. What assumptions are required for this
procedure?
b. Does the Monte Carlo simulation approach require any modification for
precedence diagramfning?
c. Does the PNET procedure require any modification for precedence
diagramming?
 APPENDIX 9-1a
THE CUMULATIVE NORMAL
DISTRIBUTION FUNCTIONt

-- for I -m r r LO)

Example: + (-3.57) = ,091785 = 0.0001785.

B Y pernlission from A. Ilald. Stntisricnl Trrbles, and Forntalas, John Wiley & Som. Inc., New York, 1952.

315
 APPENDIX 9-1b
THE CUMULATIVE NORMAL
DISTRIBUTION FUNCTl0N-f

Example: o (3.57) = .9:48215 = 0.9998215.

t By per~~liniun
from A. Hald, Stotisticnl Tables, nnd Forntulns, John Wiley h Sons, Ine., hTew York, 1952.

316
APPENDIX 9-2
USE OF
HISTORICAL DATA
IN ESTIMATING
a, m, AND b

Occasionally one may have historical (sample) activity duration data on which to
base estimates of t, and (v,)'l2, or better, to estimate a, m, and b, which when
processed in the usual manner, with equations (3) and (4), will give the desired
estimates of (v,)''~ and t,. This procedure has merit, if the following condi-
tions are satisfied.

1. The historical data are representative of the hypothetical population (Fig-

ures 9-2 and 9-8) to be "sampled" in the future for the activity in ques-
tion; that is, the activity is precisely the same, and the conditions which
prevailed during the collection of the historical data are representative of
those expected to prevail in the future when the activity in question is to
be performed.
2. The sample of historical data is of "sufficient" size. Quantitative specifica-
tion of what is "sufficient" depends on the nature of the activity in ques-
tion and the experience and abilities of the person supplying the estimates;
however, a sample of less than four or five observations would generally
not be considered "sufficient."

If the above assumptions are satisfied, estimates of a, m, and b can be obtained

from equations (8), (9), and (10) below, wherein
R = range of sample data
= largest observation-smallest observation
318 11 l Advanced Topics

Table 9-8. Constant to Convert the Range to Estimates of the Standard

Deviation

(RangelStd.
Sample Slzet Dev.) = d*tt k = 1.6/dZ

t Although this table includes samples ar small as two, one should not rely solely on tbe
sample data unless the sample size is at least four.
tt The symbol d, used here is the universal designation of this ratio, which is widely used
and tabled in 8t;ltistical quality control literature; it assumes the random variable is normally
distributed.

k = 3/d,, where d , is the statistical quality control constant tabled as

a function of the number, n, of acitivity times in the sample data.
Actually, d, is the average of the ratio R / ( v ~ ) ' / ~Values
. of k are
given in Table 9-8, and are used to compute the constants a and b.
i = arithmetic average of the sample data
Estimate of m = i (8)
Estimate of a = f - kR (9)
Estimate of b = f + kR (10)
In situations where kR is greater than f, and hence a as given by equation ( 9 )
is negative, it is suggested that the following be used.
a=O (94
b = 2f (lea)
OTHER NETWORKING
SCHEMES

The graphical methods of drawing networks introduced in Chapter 2, and used

throughout this text, were variations of two basic methods of expressing a proj-
ect plan, that is, using activities-on-arrows (arrow diagrams) or activities-on-
nodes (node or precedence diagrams). The original developers of PERT and
CPM both utilized arrow diagrams based on the simple finish-start precedence re-
lationship. It was also pointed out in Chapter 2 that for every such arrow dia-
gram, an equivaleat node diagram could be drawn by simply reversing the roles
of the arrows and nodes. Using this scheme, which is growing in popularity, the
nodes represent the activities and the arrows are merely connectors denoting
finish-start precedence relationships. The principal advantage of this basic node
scheme is that it eliminates the need for special dummies to correct false depen-
dencies. For this reason, it is easier to learn to draw the networks, and computer
processing is somewhat simplified since inputloutput lines devoted to dummy
activities are unnecessary.
The final network development in Chapter 2 was the introduction of the
precedence diagramming version of node diagrams. This scheme elttended the
simple logic of the basic node scheme. It permitted the start and/or finish of an
activity to be tied into the start and/or finish of other predecessor or successor
activities by the use of 4 types of precedence relationships. The addition of
leadllag times to these precedence relationships gave added flexibility to this
method of network diagramming.
The basic arrow diagram and node diagram schemes have, indeed, served quite
319
320 11 I Advanced Topics

well in modeling project plans. However, at the cutting edge of project planning
methodology, there are situations that still cannot be adequately modeled by
these basic schemes. Arrow and node diagrams are based on deterministic net-
work logic. That is, it is assumed that every path in the network is a necessary
part of the project; there are no optional or alternative paths. However, we
know that in some types of projects there is some uncertainty as to just which
activities will be included. In research projects in particular, several different
outcomes of the project may result, depending upon the outcomes of certain
chains of activities. A network that shows only one plan with one possible out-
come may not represent adequately the true nature of the project. The flexibil-
ity desired here is called probabilistic branching, where only one of several suc-
cessor activities leaving a burst event are realized. Another important deficiency
of basic network schemes is that they do not permit the incorporation of loops
in the networks.
A simple example of an application of looping and probabilistic branching
would be the activities following a "test" activity. If the test result isacceptable,
then the branch to the next phase of the project would be followed. However,
if the test result is unacceptable, then an alternate branch is followed which
forms a loop to correct the unacceptable performance, and eventually leads back
to repeat the same test activity.
The use of branching and looping in this situation could be avoided in simple
arrow or node diagrams by incorporating testing, rework if needed, retesting,
etc., all in one activity. These two alternatives are shown in Figure 10-1. If the
PERT system of time estimation were utilized in this situation, the uncertainty
in the Testing and Rework (10-20) activity could be accounted for by assigning

Network with probabilistic branching and loops

(See Table 10-1 below for an explanation of the node symbols used here.)

Acceptable I Tn /

t Rework (Recycle loop1

Eauivalent simole arrow diaoram

Testing
Initial To
project . and rework
) conclusion
activities of project

Figure 10-1 Example o f probabilistic branching and looping, with an equivalent single ar-
row diagram activity 10-20.
 Othar Networking Schemes 321

Initial
Evaluation of
alternative engine
a Drop project

Proceed to develop
solid fuel engine
project
activities
Proceed to develop

t - Rework (Recycle loop)

liquid fuel engine

(See Table 10-1 below for an explanation of the node symbols used here.]
Figure 10-2 Example of probabilistic branching and looping--no equivalent arrow diagram
possible. (See Table 1 0 1 below for an explanation of the node symbols used here.)

an appropriate range of activity duration times, with the optimistic time corre-
sponding to an acceptable test on the first try, and a pessimistic duration corre-
sponding to the time required to traverse a maximum number of recycle loops
that might reasonably be expected to occur in practice.
Now, consider a situation such as shown in Figure 10-2. Here, the testing or
evaluation activity (10-20) has a probabilistic successor event (20) that branches
to one of 4 possible successor activities, one of which forms a loop leading back
to the same evaluation activity. Note also, in this case, that there are several dif-
ferent project end events, only one of which can ever be realized. The true es-
sence of this project plan, and the evaluation of alternative project policies,
could not be captured by basic arrow or node diagrams, such.as was done in
Figure 10-1. The generalized arrow diagram which incorporates the branching
and looping shown in Figures 10-1 and 10-2 was developed by Alan Pritsker and
is called CERT, the acronym for Graphical Evaluation and Review Technique
(References 13-18).

DEVELOPMENT OF GERT

One of the earliest extensions of the basic arrow and node diagrams was sug-
gested by Freeman6 and dealt with the portrayal of alternative networks to rep-
resent a project. Eisner2 generalized this concept by including a new node type
for branching events to graphically portray alternative paths in a network. The
probabilities assigned to these paths add to one, and they are estimated by the
project planners along with the activity duration times on the basis of judgment
and experience with similar circumstances. A significant breakthrough in gen-
eralizing the structure of activity networks was then made by Elmaghraby3*4
when he defined three types of nodes: AND nodes; EXCLUSIVE-OR nodes;
and INCLUSIVE-OR nodes. The initial GERT developments were made by
pritsker'3.14, 15.17 by building on Elmaghraby's node definitions.
 The first research activities on GERT were performed at the Rand Corpora-
tion whiie developing procedures for automatic checkout equipment for the
Apollo Program. Pritsker subsequently worked with many groups in leading up
to the current program called Q-GERT. Notable from a project management
point of view was the addition of costs to the usual time analysis of networks.
Arisawa and Elmaghraby modeled costs by considering setup costs and variable
costs that increase in time for each project activity. P-GERT was developed by
PritskerI6 to handle node diagram notation. Finally, HebertS developed the pro-
gram called R-GERT which gives the basic outputs of a simulation of a PERT
network (as discussed in Chapter 9) with the consideration of resource con-
straints. That is, the program gives the earlyllate startlfinish times and criti-
cality indices for each project activity, with the condition that an activity
cannot be started until the stated resources required by the activity become
available.
In the next section ,GERT nomenclature will be introduced and its use will be
illustrated for a simple project of reviewing a paper for a technical journal. Then,
a detailed example of Q-GERT will be presented which deals with an industrial
sales negotiation process.

GERT NOMENCLATURE

One form of basic GERT nomenclature is shown in Table 10-1. The basic arrow
diagram logic of PERTICPM is embodied in the circle node shown in the upper
left-hand quadrant of this table. The other three node symbols in this table are
GERT logic extensions which permit looping, branching, and multiple project
end results.
An application of this notation is given in Figure 10-3, which treats the pro-
cess of reviewing an article submitted for publication in a professional journal.
The network contains one initial event and two terminal events; the latter have
the symbol for infinity (m) in the lower quadrant to denote that they can only
be realized once. These events are halfnodes because the initial event has no in-
puts, and the terminal events have no outputs. The network chart is essentially
self-explanatory, which is one of the advantages of this nomenclature. This
could be called a multiple reviewer policy which requires a favorable majority to
accept the article for publication. The network involves branching at nodes 4,5
and 8 t o denote the appropriate reviewer decision. Also, there are two loops
back to node 6 from nodes 8 and 13, and two end nodes denoting acceptance or
rejection of the article. Of particular note is node 13 which is realized only if
the first two reviewers do not agree (one accept and one reject); it triggers off a
recycle loop to node 6 to select an alternate reviewer.
The diagram in Figure 10-3 could, by itself, be quite useful as a general tool of
 Other Networking Schemer 323

Table 10-1. A Form of Basic GERT Notation

\ Output
Input

e
Deterministic,D

a @
Protmti'istic, D 0 @
N Denotes the node identification number.
AND The A N D node will be released only if all activities
(arrows) leading into the node are realized. The time
of realization is the largest o f the completion times o f
the activities leading into the A N D node lbasic PERTl
CPM input node logic).
F Denotes the number o f predecessor activities that
must be completed for the First realization o f the
node, and S denotes the number required for Subse-
quent realizations.
OR The OR node will be realized the first time when any
F o f the total number o f activities leading into the
node are completed. I f the node is contained i n a
loop, then the node can be realized the second and
all subsequent times when any S o f the total number
of activities leading into the node are completed; usu-
ally F 2 S.
Deterministic A l l activities emanating from the node are subse-
quently taken i f the node is realized (basic PERTl
CPM output node logic).
Probabililic Only one activity emanating from the node is taken i f
the node is realized. The sum o f the probabilities as-
sociated with each of the output activities equals one.

systems analysis, particularly where the problem is more complex than shown
here. It could be used to show redundancies, inefficiencies, and inconsistencies
in the policies or procedures in question. It could also be useful in speeding up
action at each branching node in the network, since the requirements and impli-
cations of each alternative are clearly stated. If additional data is added to this
network, the use of the Q-GERT simulation program could provide estimates of
the probabfiities of realization of each of the end nodes, and time and cost dis-
tributions for the realization of any selected nodes, considering resource con-
straints if desired. An example of this type will be presented in the next section.
 Author mails article Log article, select
to journal ed. reviewer and mail

Figure 10-3 Generalized IGERT) model of a sequential article review process. (See Table 10-1 for an explanation of the node symbols used
hers.)
 Other Networking Schemes 325

AN APPLICATION OF Q-GERT

Given below is an application of Q-GERT by Moore and Clayton12*, to model

the costs associated with an industrial sales negotiation process. A major oil
company wanted to buy a new gasoline plant as a completely installed package.
Systematically, the purchasing agent and his buyers began to inform a select
number of prospective vendors of the proposed gasoline plant and invited these
vendors to negotiate for the job. Figure 1 0 4 is a Q-GERT network model which
could have been used by one of the vendor firms. The activity numbers are ref-
erenced to Table 10-2 which contains a description of each activity along with
the time distribution, cost estimates and branching probabilities for that activ-
ity. Fixed and variable costs are associated with the negotiation activities. A
fixed cost is charged each time an activity occurs while variable costs accumu-
late linearly over the period of time required to complete the activity. Each ac-
tivity will now be discussed with activity numbers given in brackets after the
description.
The sales negotiation process involves the initial contact by company salesmen
[I]. A report to the marketing vice president and t o the president is then made
[2,3]. For this company, a decision to perform preliminary analyses is standard
for the large type of project being considered. As will be seen, an evaluation of
this policy could be cost effective. Next, on the Q-GERT network, five parallel
studies are performed beginning at node 5. These involve engineering [4] , pro-
duction [5], financial [6], marketing [7] and purchasing [a]. Except for the
production report, the evaluation of the reports is performed at the nodes fol-
lowing the activities, as represented by activities 9 and 10 and 13 through 18. If
any of the reports is negative, the project is not negotiated further and the net-
work terminates at node 22. Node 22 is repeated three times in Figure 10-4 for
graphical convenience. The evaluation of the production schedule is slightly
more complicated in that the possibility for production subcontracting is investi-
gated [ I l l . An unfavorable production subcontract could also occur [19]
which would lead to termination of the negotiation process. When all five re-
ports are favorable, node 12 is released. Node 12 is a statistics node and the
time at which node 12 is released will be collected. A user function is employed
at node 12 to collect cost data with regard to the costs of the sales negotiation
process up to the time at which five favorable reports are obtained. Based on
the five reports, corporate level planning conferences are held [21]. Further in-
formation is then sought of a marketing and engineering nature [22, 231. If
either of these calls results in unfavorable information, the sales negotiation pro-
cess is again terminated [24, 261. If positive information is obtained from the
calls [25, 271, marketing negotiation plans and engineering design plans are for-

*This example is taken from the article "Sales Negotiation Cost Planning for Corporate
Level Sales," by M. M. Bird, E. R. Clayton, and L. J. Moore, J O U ~ MMarketing
~O~ (April),
Vol. 37, pp. 7-13, by permission of the publisher, the American Marketing Association.
326 11 I Advanad Topics

*Numbers in squares below each activity are used t o identify the activities listed in Table
10-2; Activity 0, for example, could have alternatively been called 2-3,using the node num-
bers for identification.
**Node 22 appears three times i n this network t o facilitate drawing the network; they
should be interpreted as the same node.
Figure 10-4 Q-GERT diagram of industrial sales negotiation model.
 Other Networking Schemes 327

mulated [28, 291. Based on these plans, corporate level strategy is developed in
a conference [30]. This strategy is implemented in a negotiation conference
with the buying firm 1311. The results of the negotiation conference are: no
sale [32] ; contract awarded [33] ; or further negotiations requested by the
buyer involving plan modifications [34]. Marketing and engineering modifica-
tions are then performed [35, 361. A corporate level meeting to evaluate and
reconcile the modifications is then held [37]. Following this, a return is made
to node 18 of the Q-GERT network where another negotiation conference with
the buying firm is held [31]. At node 23, a successful sale has been made.
Total costs when a successful scale is made are collected at node 23. Total costs
of the negotiation when a sale has not been made are collected at node 22.
The GERT network model for the sales negotiation process allows an analyst
to include probabilistic outcomes as modeled by the probabilistic branching at
nodes throughout the network. In addition, recycling of activities is permissible
as illustrated by the loop of nodes 18, 19,20 and 21.
Nodes 22 and 23 are of special interest because they are terminal; node 12 is
also singled out as a milestone because it denotes that all preliminary reports are
favorable, and the final phase of the project is initiated. The GERT simulation
program is instructed to collect statistics on these nodes, and for this reason, spe-
cial notation is required. For example, node 12 contains the symbols 5/m, L/F,
1, UP, 38, 12. It is read as follows. First, 5/00 denotes that 5 activities (favor-
able reports) must be completed to realize this node the first time, whiie mde-
notes that this node cannot be realized more than once. (Nodes 18, 19,20 and
21 are nodes that can be realized more than once because they are contained in a
network loop; in these cases, w is replaced by the appropriate integer 1 or 2).
The L/F symbol indicates that the time completion of the Last (fifth) activity
and the First activity are kept at this node. (Note, only the First activity com-
pletion times are collected at the other terminal nodes 22 and 23, because they
are realized as soon as the First incoming activity is complete.) The last 4 entries
indicate that on node 12, user function OJF) number 38 is used to compute at-
tribute 1 which is cumulative project costs.
The main GERT program handles project time analysis and all computer out-
puts. It contains 145 lines (FORTRAN language), 1 1 1 of which deal with data
describing the 37 network activities. The cumulative cost analysis is introduced
by the User Function, which in this case is two subroutines containing 38 lines
of computer code, half of which are cost data. The User Function in this case
assumes that each activity has an associated fixed setup cost, and a variable cost
that is linear in time. Other functions of time could be used by changing one
FORTRAN statement. If node 22 (no sale) is realized, the User Function ad-
ministers company policy which requires that all work on the project is halted.
Thus, the program stops all activities so that their remaining costs are not in-
curred and reflected in the computed total project cost.
 Table 10-2. Activity Descriptions for Sales Negotiations Network 8
-
Activity Time in hours Cost in dollars* Proba-
No. Activity Description Distribution Mode Min hlax (a) (b) bility 9
LY
1 Sales call by company salesman Beta 4.0
2 Sales report to marketing vice president Beta 1.o
3 Marketing vice president reports to president Beta 0.50
4 Preliminary engineering report Beta 40.0
5 Preliminary production report Constant N/A
6 Preliminary financial report Constant N/A
7 Preliminary marketing report Constant N/A
8 Preliminary purchasing report Beta 40.0
9 Negative engineering report examined Constant NIA
10 Favorable engineering report examined Constant NIA
11 Production subcontracting investigated Beta 40.0
12 Favorable produclion report examined Constant N/A
13 Favorable financial report examined Constant NIA
14 Negative financial report examined Constant N/A
15 Favorable marketing report examined Constant N/A
16 Negative marketing report examined Constant NIA
17 Favorable purchasing report examined Constant NIA
18 Negative purchasing report examined Constant N/A
19 Unfavorable production subcontract examined Constant NIA
20 Favorable production subcontract examined Constant N/A
21 Corporate level planning conference Constant N/A
 Sales call by marketing vice president and salesman Constant NIA
Engineering call by engineering vice president Constant N/ A
Unfavorable sales call conference Beta 18.0
Favorable sales call no conference Constant NIA
Unfavorable engineering call conference Beta 16.0
Favorable engineering call no conference Constant N/A
Marketing negotiation plan formulation Beta 40.0
Engineering design plan formulation Beta 160.0
Corporate level strategy conference Beta 8.0
Negotiation conference with buying fum Beta 16.0
No sale Constant NIA
Contract awarded Constant N/A
Modifications requested by buyer Constant NIA
Modification of marketing negotiation plan Beta 20.0
Modification of engineering design plan Beta 80.0
Corporate meetina to reconcile modifications Beta 4.0
*Column (a) in the fixed cost for the activity while column (h) is the variable cost for the activity assumed to be linear over time.
330 11 I Advanced Topics

RESULTS OF COMPUTER SIMULATION

The Q-GERT model of the industrial sales negotiation process was simulated 500
times. Summaries of the time and cost results are presented in Figures 10-5 and
106. The estimates of the probability of losing a sale, node 22, is 0.844, that is,
over 84 percent of the negotiations end in failure.
Looking at node 12, favorable reports received, it is seen that approximately
43 percent of the potential projects result in a l l five reports being favorable.
Thus, 57 percent of the sales negotiations are turned down for internal reasons.
The time estimates indicate that it takes over 38 days to decide that negotiations
should be carried beyond the internal report phase. The cost data indicate that
it costs over $3050 when all favorable reports are obtained. This information
can be extremely useful for an analyst who is attempting to improve the sales
negotiation process, as it provides trade-off data regarding the possibility of in-
creasing the probability of favorable reports versus the time and costs required
to obtain the favorable reports. By developing alternative networks up t o node
12, such trade-offs can be made. The analyst should attempt to develop proce-
dures for detecting when unfavorable reports will be issued. Possibly sequential
reporting of the engineering, production, financial, marketing and purchasing re-
ports should be made. Since engineering and marketing have the highest prob-
ability of issuing a negative report, possibly these two activities should be per-
formed prior to the other reports. If this is done, the time required to reach the
decision with regard to the preliminary reports will be extended.
The summary statistics for a lost sale indicate that it takes 62 days on the av-
erage to make this determination. Since node 22 can be reached from many
points in the network, this time should have a wide variability which is the case
as indicated by its standard deviation of 56.4 and its range of 23 to 288. The av-
erage cost associated with a lost sale is $4944. This indicates that when the proj-
ect fails more money is expended in obtaining favorable reports than is put into
the negotiation effort after the reports are obtained. When the project succeeds,
which occurs only 15.6 percent of the time, the negotiation process takes a
lengthy 182 days and costs on the average $1 1,947. Other statistical quantities
concerning the time and cost when the negotiation is successful are shown in
Figures 10-5 and 10-6.
The time and cost histograms associated with node 22, lost sale, are presented
in Figure 10-7. These histograms illustrate that the distribution function as-
sociated with node 22 has discrete breaks due to the different paths that can
be used to reach node 22. For example, failed reports occurred on 283 runs.
((1 - .434) X 500 = 283). The histogram for node 22 shows 283 values in the
range 20 to 50; hence, this cluster of values is associated with failed reports. The
other values in the histogram for node 22 are for the times when a lost sale oc-
curred after favorable reports were received.
 **FINAL YESJLTS FOR 5 0 0 SXHULATXONSD*

*+AVE3ASE NODE S T A T l S T I C S * *

NODE LAIKL PRJj43lLITV A#€. STO.OEY. S O OF P V E NO O f WIN. MAX. STAT

00s. TVPE

22 LOST-SAL -84CO 62.1155 5b.4425 2.7576 422. 22.6712 287.7906 f

23 SUCCESS .15b0 182.Olbc *9.1430 5.4512 7 8 125.7301 425.1386 F
12 FAV-UPTS .C3*0 30.103C 5.6356 -31116 217. 28.5247 50.9886 f
Figure 10-5 Summary report for times associated with sales negotiation model.

P
+ + U S E 9 S T A T I S T I C S FU9 V A S I P 6 L E S BLSED ON OBSEQVLTION AT T I M E * 3 0 . 0 9 5 I N RUN 500** 33
A VE S T J DEII S O OF A V E n~hxnun n ~ x ~ n g w 06s 5'
(D
V1
FAV-RPTS 3053.6797 141.4592 9.5036 2 760.4418 3469.0bO2 217 %
LOST-SAL 4943.8782 3330.3886 162.1201 2475.4571 17774.6202 422 I
8
SUCCESS 11946.3841 27JO.lble 304.1299 95ZC.0147 25230.0935 78
Figure 10-6 Summary report for costs associated with sales negotiation model.
W
9
332 11 I Advanced Topics

r Sf., .1*10i*L3 FOR "DOC 2l

Loll-SlL

UPPE?
13>*0 0' CELL 8

I
I....!....!.
21
... b I b0
1....1...I....I....l....I....I....I
I* LOO

I
I........................ I

:.....
I......... c I
C I
C I
I... C I
I C I

::...
I C I
C I
t
I.. C I
I C I
1. C I
I. C
1 C
I C
I C
I C
I C
I C
T..I*..l..I..I..I..I..I..l..1..I

Figure 10-7 Histograms of times and costs associated with unsuccessful sales negotiation.

Histograms for the time and costs associated with a successful sales negotia-
tion were also developed, but are not presented here. These histograms could be
used to monitor company time and cost data associated with successful sales
negotiations. They would indicate that over 20 percent of the time, the total
sales negotiation cost will be greater than $14,000. Thus, we can expect 1 time
out of 5 to have this high cost, given that the sales negotiation pattern follows
the one described by the Q-CERT model. By making such comparisons, the
company can maintain some control over the costs involved in their sales nego-
tiations activities.

SUMMARY DISCUSSION OF GERT

The above example is illustrative of the use of GERT in project management. In

this case, the probability outputs in Figure 10-5 could have been computed
 Other Networking schsmea 333

analytically using the basic laws of probability. However, this is not true for the
time and cost statistics in Figures 10-6 and 10-7. In a more complex hardware
development project by Moderia, it was found to be difficult to intuitively pre-
dict even the direction of the effect on time and costs of policy changes in a
project plan. One change led to a reduction in both time and cost. GERT simu-
lation is particularly useful in such cases.
Some 13 years have passed since the previous edition of this book in 1970.
At that time it was expected that GERT would prove to be a useful tool in proj-
ect management. The literature of the 709 attests to the validity of this forecast.
For example, Vanston19 used GERT to analyze the effects of various funding
and administrative strategies on nuclear fusion power plant development. Halpin7
used GERT to investigate the use of simulation networks for modeling construc-
tion operations. Another important application in this area (see reference 5) in-
volved risk analysis of the construction schedule of the Alaskan pipeline. The ef-
fects of weather conditions on construction activities was modeled within the
Q-GERT framework to estimate the distribution of both project time and costs
performance. Many other applications could be cited.
It has been pointed out above that PERT network logic is a subset of GERT
logic. Hence, the latter can be used t o simulate PERT networks t o obtain the
probability distribution of project duration and activity criticality indices. At
present, this application would require separate computer runs t o produce a
GERT simulation output and a conventional PERT output. A number of re-
searchers (Wolfezo and Hebert9) are developing programs to produce both these
outputs from a single computer run. When the objective is both project planning
and control, this is the desired mode of operation and the development of such
programs should provide a marked increase in the use of GERT in this area of
application.
The most recent development in this area of stochastic networks by Moeller
and Digman", has been called VERT, the acronym for Venture Evaluation and
Review Technique. It is a computer-based mathematical simulation network
technique designed t o systematically assess the risks involved in undertaking a
new venture and in resource planning, control monitoring and overall evaluation
of ongoing projects, programs and systems. It involves even more User Function
types of interaction than GERT, and is perhaps a logical evolution t o embrace
the somewhat elusive third parameter of the time/cost/quality of performance
triumvirate.

REFERENCES

1. Arisawa, S. and S. E. Elmaghraby, "Optimal Time-Cost Trade-offs in GERT

Networks," Management Science, Vol. 18, No. 11, July 1972, pp. 589-599.
2. Eisner, H., "A Generalized Network Approach to the Planning and Sched-
uling of a Research Project," Operations Research, Vol. 10, No. 1, 1962,
pp. 115-125.
334 11 I Advancad Topics

3. Elmaghraby, S. E., "An Algebra for the Analysis of Generalized Activity

Networks," Management Science, Vol. 10, No. 3, 1964, pp. 494-514.
4. Elmaghraby, S. E., Activity Networks: Project Planningand Control by Net-
work Methods, John Wiley & Sons, New York, N.Y., 1977.
5. Federal Power Commission Exhibit EP-237, "Risks Analysis of the Arctic
Gas Pipeline Project Construction Schedule," Vol. 167, Federal Power Com-
mission, 1976.
6. Freeman, R. J., "A Generalized PERT," Operations Research, Vol. 8 , No. 2,
1960, p. 281.
7. Halpin, D. W., "An Investigation of the Use of Simulation Networks for
Modeling Construction Operations," Ph.D. dissertation, University of Illi-
nois, 1973.
8. Hebert, J. E. 111, "Critical Path Analysis and a Simulation Program for
Resource-Constrained Activity Scheduling in GERT Project Networks,"
Ph.D. dissertation, Purdue University, 1975.
9. Hebert, J. E. 111, "Project Management with PROJECT/GERT," ORSA/
TIMS Joint National Meeting, November 1978, Los Angeles, Calif.
10. Moder, J. J., R. A. Clark, and R. S. Gomez, "Applications of a GERT Simu-
lator t o a Repetitive Hardward Development Type Project," AIIE Transac-
tions, Vol. 3, No. 4, 1971, pp. 271-280.
11. Moeller, G. L. and L. A. Digman, "Operations Planning with VERT," Opera-
tions Research, Vol. 29, No. 4 , July-August 1981, pp. 676-697.
12. Moore, L. J. and E. R. Clayton, Introduction to Systems Analysis with
GERTModeling and Simulation, Petrocelli Books, New York, N.Y., 1976.
13. Pritsker, A. A. B., "GERT: Graphical Evaluation and Review Technique,"
The Rand Corporation, RM-4973-NASA, Santa Monica, California, April
1966.
14. Pritsker, A. A. B., and W. W. Happ, "GERT: Graphical Evaluation and Re-
view Technique, Part I. Fundamentals," Journal of Industrial Engineering,
Vol. 17, No. 5, 1966, pp. 267-74.
15. Pritsker, A. A. B., and G. E. Whitehouse, "GERT: Graphical Evaluation and
Review Technique, Part 11. Applications," Journal o f Industrial Engineer-
ing, Vol. 17, No. 5, 1966, pp. 293-301.
16. Pritsker, A. A. B., The Precedence GERT User's Manual, Pritsker & Asso-
ciates, Inc., Lafayette, TN, 1974.
17. Pritsker, A. A. B., and C. Elliott Sigal, Management Decision Making: A
Network Simulation Approach, Prentice-Hall, Inc., Englewood Cliffs, New
Jersey, 1983.
18. Pritsker, A. A. B., Modeling and Analysis Using Q-GERTNetwork, Halsted
Press Books (John Wiley & Sons), New York, N.Y., 1977.
19. Vanston, J. H., Jr., "Use of the Partitive Analytical Forecasting (PAF) Tech-
nique for Analyzing of the Effects of Various Funding and Administrative
Strategies on Nuclear Fusion Power Plan Development," University of
Texas, TR ESL-15, Energy Systems Laboratory, 1974.
20. Wolfe, P. M., "Using GERT as a Basis for a Project Management System,"
ORSAITIMS Joint National Meeting, Los Angeles, California, November
1978.
 Other Networking Schemes 335

EXERCISES

1. Construct a table containing a list of criteria that you feel could be used t o
measure the utility of networking schemes. On the other axis list all the net-
working and project charting schemes you are familiar with, including hybrid
forms of bar charts, milestone charts: or others. Then assign a rank or other
value score t o each scheme and criterion and summarize the score of each
scheme. Which scheme received the highest score?
2. Compare the results of your scoring in the exercise above with those of other
students and, if practical, with the opinions of persons experienced with a
variety of networking schemes. As a result of these comparisons, comment
on whether it is practical t o evaluate networking schemes on a technical, ob-
jective basis.
3. The techniques of resource allocation (Chapter 7) are applicable to certain
types of job-shop scheduling problems, especially where the sequence of jobs
is fixed and the problems are reduced t o questions of the loading of certain
facilities or pools of resources and the avoidance of project (or job) delays.
Discuss how generalized network concepts may provide solution methods t o
a wider class of job-scheduling problems.
4. Draw a GERT network for the following procedure used by a university
patent committee.
1. Inventor submits invention t o patent committee.
2. Above (1) initiates two concurrent activities:
2.1. Technical or commercial reviews by experts; and
2.2. Legal review of inventor's patent liability to contractors and
university.
3. Study of legal and technical reviews.
4. The study in (3) leads to
4.1. Favorable evaluation, or
4.2. Unfavorable evaluation
5. Favorable evaluation in 4.1 leads t o
5.1. Submission of invention t o university patent attorney; or
5.2. Submission of invention to outside patent corporation.
6. Activity 5.1 leads t o
6.1. University marketing of patent; or
6.2. University drops invention because of lack of patent protection.
7. Activity 5.2 leads t o
7.1. Outside patent corporation rejects invention; or
7.2. Outside patent corporation seeks patent.
8. Activity 7.1 leads t o
8.1. Resubmission by university patent committee of patent t o outside
patent corporation for reconsideration; or
8.2. University drops invention.
9. Activity 7.2 leads t o
9.1. Patent denied so university drops invention; or
9.2. Patent acquired so outside patent corporation markets patent.
336 11 / Advanced Topics

10. Unfavorable evaluation in 4.2 leads t o

10.1. Submission of invention to outside patent corporation; or
10.2. University drops invention.
5. Draw a GERT network for a general R&D process which includes the follow-
ing activities or nodes: start project; problem definition; research activity;
evaluate solutions; solution unacceptable more research; solution unaccept-
able redefine problem; project washout; develop prototype; redevelop proto-
type; solution implementation; project successful completion.
6. Consider drawing a GERT network for an embellishment of Problem 5. Sup-
pose there are 4 projects and two research teams. Team 1 is t o start on proj-
ect 1, and team 2 on project 4. When a team completes a project (for any
reason), it immediately moves on t o a project that has not yet started. Team
1 is always t o select the lowest project number available, while team 2 selects
the highest project number available. The simulation is t o continue until all
4 projects are complete. Note, at the end of the simulation period, one team
will be idle, rather than helping the other team that is still working.
COMPUTER PROCESSING

In Chapter 4, manual methods of making the basic critical path scheduling com-
putations were presented. This chapter will describe how the samecomputations
can be made by computers. The computer programs for CPM and PERT compu-
tations are available to anyone who has access to almost any general purpose
business computer (including so-called minicomputers and microcomputers).
The treatment in this chapter begins with what the typical critical path soft-
ware is and how one can make use of it. The actual software that one may find
will vary somewhat in capacity, speed, features provided, and even sometimes in
the computational results. Consequently, this text does not attempt to explain
how each available program works. Rather, it is a treatment of how they all
work in general, and what to look for in the way of comparative features.
The basic critical path scheduling computations are emphasized in this chap-
ter. The roles of the computer in the more advanced topics are mentioned here
but are covered more fully in the appropriate chapters.

ADVANTAGES OF COMPUTERS

Computer processing of networks offers several advantages over manual calcula-

tions. Among these are:

1. The project activities can be easily sorted and listed in a variety of useful
ways, such as by float (critical path listing), by early or late start dates
338 11 I Advanced Topics

(showing what needs to be done next), by responsibility codes (providing

lists for each supervisor or subcontractor), by activity numbers, and by
other keys or combinations of keys. These sorted lists help simplify the
management tasks.
2. Once the network is stored in the computer the project schedule can be
easily updated, providing new listings based on actual progress on the proj-
ect or changes in estimates. This is a key advantage in making critical path
methods useful throughout a project's life cycle.
3. Certain analyses that go beyond the basic schedule calculations, such as
time-cost trade-offs (Chapter 8), resource allocation (Chapter 7), and cost
control (Chapter 5) are really practical only with the aid of computer
programs.

In spite of these advantages, however, a computer is certainly not required or

necessarily desired for all critical path applications. Computer processing does
involve extra time and cost to transfer the network into a computer input for-
mat, and in many cases the extra cost is not justified. In other cases a suitable
computer is simply not available.

WHEN A COMPUTER IS NEEDED

In deciding whether to process a particular network by manual calculations or

by computer, the following factors should be considered:

1 . Network size. The larger the network, of course, the more likely that
computer processing will be needed. However, there is no specific number
of activities that would call for a computer, due to the influence of the
other factors listed below.
2. Computer avm'ZabiZity. Both a suitable computer and adequate critical
path software must be available at a reasonable cost.
3. Expected frequency of updating. If the network is primarily a planning
tool and is not likely t o be revised and updated throughout the project,
computer processing may not be worthwhile, even if the network is large.
4 . Desiredoutput listings. If printed listings of the network activities are de-
sired, selected, and sorted by float, responsibility, start dates, activity num-
bers, etc., and if these lists are desired for each updating calculation on the
project, then computer processing is the most practical answer.
5. Advanced analyses. In most cases computers and the appropriate soft-
ware are required for timecost trade-off, resource allocation, PERT statis-
tistics, and cost control analyses.
6 . Network format. When the precedence network format is used, manual
calculations of the basic schedule are complex, making the use of com-
puter processing decidely advantageous.
 Computsr Processing 339

Each potential user must consider these factors in relation to his project and
the particular program available. A few hypothetical examples of such consider-
ations are given below. These examples are not offered as firm guidelines for the
types of situations described, but rather as indications of the considerations that
should influence the decision of whether or not to use a computer.

CASE 1 . Heart Surgery. A network of 75 activities has been prepared for the
purpose of planning a complex surgical operation. The network will
not be used after the plan has been worked out. Recommendation: In
this case a manual computation and a time-scaled network would be
the best means of working- out a well-coordinated plan.
CASE 2. Promotional Campaign. A project to introduce a new consumer prod-
uct involves about 80 advertising, manufacturing, distribution, and
sales activities. The network is to be used for initial planning and coor-
dination throughout the six-month project in order to assure proper
timing of each phase. All the supervisors have had a short course in
CPM. Management intends to update the network twice a month and
distribute copies of the results to each of the key supervisors involved.
Recommendation: In this case the frequency of updating indicates that
computer processing would be faster and more economical than man-
ual updating. Also, the computer can sort the output by responsibil-
ity, so that each supervisor can get an extracted report on only his
activities.
CASE 3. Construction of a Lazge Building in a Remote Location. A five-story
building is to be constructed in a small city where no critical path com-
puter program is available. The local contractor is interested in apply-
ing CPM in planning the project, though, because this will be his first
multi-story building project. The first draft of his network contained
270 activities. Recommendation: The manual computation of this net-
work will be well worthwhile, perhaps saving the contractor a great
deal of cost and confusion over the re-use of concrete forms, the phas-
ing of carpentry, steel, electrical, and other trades, and by helping to
solve other problems of coordinating multi-story construction that he
has not yet experienced. (See also Exercise 7.)
CASE 4. New Plant Start-up. This project includes construction of a plant,
delivery and installation of machinery, recruiting and training a new
labor force, and starting production. The key supervisors for the con-
tractor, machinery suppliers, training and production departments have
had no CPM training, and management considers it impractical to un-
dertake such training at this time. The project manager's network con-
tains 210 activities, and he intends to use it for schedule control during
the project. A good CPM program is available. Recommendation: The
pros and cons of using a computer in this case are not strong either
340 11 / Advanced Topia

way. The computer reports may be confusing and useless to supervi-

sors not trained in how to read them. On the other hand, experience
has shown that machinery suppliers and subcontractors tend to place
greater effort on schedule control when the customer shows them
periodic computer reports indicating that they will delay the project
unless certain tasks are completed by certain dates. Here the psycho-
logical factors are more significant than the time or costs of process-
ing. The decision could go either way, depending on the individuals
involved.

COMPUTER SOFTWARE

The term computer software refers to the recorded instructions or programs

necessary to make a computer perform the functions desired, including the read-
ing of input data, testing for errors, calculating, sorting and printing out reports,
etc.
We shall use the term critical path software to describe the computer pro-
grams that perform at least the basic forward pass, backward pass and float com-
putations associated with CPM or PERT networks. Usually the software is a se-
ries of programs (or software package) designed to operate on a particular model
of computer but will accept any network data, provided the network data follow
certain rules and do not exceed the capacity of the software. Thus new software
does not have t o be developed for each network.
The software is usually stored on a magnetic disk or tape.
Each software package has a booklet of instructions for the user, explaining
how to prepare the network data for input, how to select various processing op-
tions, and how t o read the output reports. By following these instructions care-
fully one can translate a network into the required input format and let the com-
puter do all the critical path processing.
After the initial computation has been made, the computer may store the
user's network on a magnetic tape or disk t o await updating computations. A
schematic diagram of these operations is given in Figure 11-1. At the time of up-
dating, the user provides input data on the progress of the project and changes in
the network. These data are then processed against the initial network the com-
puter has stored. A schematic diagram of this updating function is shown in
Figure 11-2.

INPUT FOR INITIAL COMPUTATION

To illustrate how a typical critical path software package may be used, let us
take the market survey network in Figure 11-3 and prepare it for computer pro-
cessing. (In this case we will be employing the arrow scheme of networking. In-
 Computer Processing 341

network data

(Keyboard entry
or punched cards1
lnittal CPM
computations

lcomputerl

- Initial
CPM
reports
(tape or disk)

Figure 11-1 Schematic flow diagram of initial CPM processing.

Updating CPM
computations

- Updated
CPM reports

lnew file)

Figure 11-2 Schematic flow diagram of updating computation.

342 11 I Advanced Topics

Plan Conduct Analvze

questionnaire

Figure 11-3 Network of survey project.

put procedures for networks in the node and precedence diagramming formats
are different. Each software package contains instructions for input.)
Most input forms for arrow-scheme networks are similar to that shown in
Figure 11-4. Data related t o an activity is entered in one row of this form.
The column labeled "pred" is for the predecessor event number of the activity,
and "succ" is for the successor event number. The column labeled "t" is for the
activity time estimate. (PERT input forms have columns for three time esti-
mates.) The description column usually accepts a certain number of alphabetic
or numeric characters, which are used to describe the activity.

COMPUTER INPUT FORM

PRED SUCC t DESCRIPTION

2 3 5 HIRE PERSONNEL

1 2 3 PLAN SURVEY

2 4 10 DESIGN QUESTIONNAIRE
-- -

4 3 0 DUMMY

6 5 0 DUMMY

4 5 4 SELECT HOUSEHOLDS

5 7 15 CONDUCT SURVEY

7 8 5 ANALYZE RESULTS

4 6 5 PRINT QUSS QUESTIONNAIRES

3 5 7 TRAIN PERSONNEL

Figure 11-4 Basic input for critical path computer programs.

 Computer Proceaing 343

The way the form is fdled out illustrates one of the rules that must be fol-
lowed: all dummies must be entered into the form just as all other activities, ex-
cept that dummies carry zero time estimates. The description "dummy" is op-
tional. Note also that the activities are entered in the form in random order,
which is permissible. (We have also assumed that the program to be used will
accept events numbered at random; otherwise numbers 3 and 4 in the network
would have to be reversed.) Other common rules of input are that each activity
must be entered, and no activity may be entered twice. Programs that have
scheduled date, calendar date, cost control, and other features have additional
columns in the input forms for certain necessary data for these features.
After the form has been filled out and double-checked for errors or omissions,
the form is given to a keyboard operator who transfers the information onto a
computer input medium such as a magnetic disk, tape, or punched cards. The
input medium is also checked for errors and then is fed into the computer sys-
tem for processing.
In addition to the data on each activity, one must also provide certain general
information and parameters for the network as a whole. These data would nor-
mally include the name of the network, whether the initial or updating computa-
tion is desired, and such other data as required by the options of the program.
For example, programs having calendar dating capability normally would require
that the network data include:

1. the network start date,

2. the network target completion date,
3. whether a 5-, 6-, or 7-day week is t o be assumed, and
4. which holidays are to be included.

The requirements for such data and their input format vary with each program,
therefore they are not illustrated here.

OUTPUT FOR INITIAL COMPUTATION

The basic output report contains the same information as given by manual meth-
ods. The typical format is shown in Figure 11-5. Note that the first four fields
contain the input data for each activity. The remaining fields provide the com-
puted results for each activity. In this example the times are expressed in units,
the unit being defined by the user. In most programs one may elect t o obtain
the output times in the form of calendar dates.
Note also that the output in Figure 11-5 is now in a particular order, although
the activities were put in at random. The order here is by float or slack, from
the lowest to the highest. This "slack sort" is one of the most commonly re-
quested sorts of the output, because it lists the most critical activities first, and
thus easily identifies the critical path, or paths.
344 11I Advanced Topics

MARKET SURVEY PROJECT INITIAL SCHEDULE

I J DUR DESCRIPTION ES LS EF LF S
1 2 3 PLAN SURVEY 0 0 3 3 0
2 4 10 DESIGN QUESTIONNAIRE 3 3 13 13 0
4 3 0 DUMMY
3 5 7 TRAIN PERSONNEL
5 7 15 CONDUCT SURVEY
7 8 5 ANALYZE RESULTS
4 6 5 PRINT QUESTIONNAIRE
6 5 0 DUMMY
4 5 4 SELECT HOUSEHOLDS
2 3 5 HIRE PERSONNEL
Figure 11-5 Basic computer output report.

INPUT F O R UPDATING

To update a critical path schedule after the project has begun and progress has
been made, the computer programs may call for the following types of data:

1. Beginning date for activities that have begun.

2. Finish date for activities that have been completed.
3. New data for activities that have been changed.
4. Complete data for activities that have been added to the network.
5. Identification of activities that have been deleted from the network.

Most updating programs utilize only start and finish dates of activities, al-
though a few will accept a "percent completion" figure.
For changed activities, normally only the changed field must be entered. This
will usually be the time estimate or the description. The I and J numbers may
not be changed, except by deleting the activity and adding a new one with the
new numbers. The I and J numbers are the keys to identification of each activ-
ity in the computer files.
Care must be taken in deleting activities, to insure that errors or gaps are not
created in the network. If activity 4-3 in the sample network were deleted the
computer would accept the input and process it, although the result would be
different than when 4-3 is included. However, if activity 3-5 were deleted (and
not replaced by one or more other activities) the network would have more than
one terminal event; some programs would halt on this condition.
In addition to the above data for the activities, updating input must include
some general project information as in the input for the initial computation. An
important fact for the updating run is the effective or "cut-off" date of the re-
port. Some computations will involve the effective date. Also the effective date
serves as the basis for certain error checks. For example, a reported actual date
(for the start or finish of an activity) may not be later than the effective date.
 Computer Processing 345

Actual Actual
Activity Start Date Finish Date Change

2
12
-
Change time estimate
to 10 days.

Figure 11-6 Progress on survey project as of the end of the twelfth day.

Update
I J DUR DESCRIPTION Code Date

1 2 1 0
1 2 2 2
2 4 1 2
2 4 2 12
4 6 1 12
3 5 10 3

Figure 11-7 Input for updating computation.

To illustrate how typical updating data are entered, consider the survey proj-
ect of Figure 11-3. Assume that at the end of day 12 progress has been made
according to the information in Figure 11-6.
Assume that no other progress dates or changes have been recorded as of the
end of day 12. These updating data may be entered on the program input form
as shown in Figure 11-7. The key to the "Update Code" is as follows: 1 = start
date; 2 = fmish date; 3 = change. Note that only the activities that are actually
involved in the updating are entered on the form.
It is emphasized that these procedures are presented only as representative
of how some programs handle initial and updating information. The particular
program that a user may obtain will probably differ from these examples in a
variety of ways. This is especially true of the updating function. Consequently,
the user must carefully study the instruction manual provided with the program
to be used.

OUTPUT FOR UPDATED SCHEDULE

To illustrate how updating input may be handled by a computer program, let us

assume that day 40 is the established target date for completion of the market
survey project, and that the computer program will make its backward pass from
this target date in each updating run. (Some programs have this feature, others
346 11 I Advanced Topics

MARKET SURVEY PROJECT EFFECTIVE DATE: 12

Est Act
I J Dur Dur Description ES LS EF L F F

1 2 3 2 PLAN SURVEY STARTED 0 FINISHED 2

2 4 10 10 DESIGN QUESTIONNAIRE STARTED 2 FINISHED 12
2 3 5 HIRE PERSONNEL 13 5 16 10 -8
3 5 10 TRAIN PERSONNEL 18 10 28 20 -8
5 7 15 CONDUCT SURVEY 28 20 43 35 -8
7 8 5 ANALYZE RESULTS 43 35 48 40 -8
4 5 4 SELECT HOUSEHOLDS 13 16 17 20 3
4 6 5 PRINT QUESTIONNAIRE STARTED 12 17 20 3

Fiwre 11-8 Update deport.

do not.) Under these conditions the computer output should be basically as

shown in Figure 11-8.
The format of the updated report in Figure 11-8 is different from that of the
initial report shown in Figure 11 -5. The actual duration ("ACT DUR") now ap-
pears for the activities that have been completed. This figure is simply the dif-
ference between the reported start and finish dates, which are shown under
the LS and LF columns, respectively. The preceding words "STARTED" and
"FINISHED" indicate that the dates that follow are the actual reported dates
rather than the computed LS and L F dates. Note also that the finished activities
no longer show a float figure, and are listed first in the float sort.
The next four activities form the critical path, which has changed since the
initial computation. The dummy activities have been automatically deleted
from this report because they no longer play a role in the balance of the net-
work. That is, the activities preceding the dummies have been completed, re-
moving the precedence-constraint purpose of the dummies. (Some programs
continue to list the dummies anyway. Other programs offer the user the option
of omitting all dummies and completed activities from update reports.)
The update report, sorted by float, provides a straightforward statement of the
project status. One sees at a glance which activities are completed and, of those
that remain, which are most critical. The fact that the project is eight days be-
hind schedule is also immediately apparent. The cause of the late condition can
be discovered by looking at the records for activities preceding the current criti-
cal path and the initial activity on the critical path. The reader will see that ac-
tivity 1-2 started on time and finished a day early. However, activity 2-3, which
begins the new critical path, did not start as early as it could have. In fact, as of
day 12 it still has not been reported started, meaning that it cannot start until
day 13 at the earliest, which is 10 days after the ES and 5 days after the LS
time. Furthermore, the estimated duratjon of activity 3-5 has been changed
from 7 to 10 days. For these two reasons, then, the expected completion date
 Computer Processing 347

Table 11-1. Optional Output Sequences ('Sorts")

Sort Key Application

Activity Number Enables quick location of information about indi-

(I-Jor node numbers) vidual activities. Useful i n cross-referencing the
graphic network with the computer output and for
error-checking.
Float (or "Total Slack") Organizes the activities into paths, with the most
critical paths listed first. (Figure 11-8)
Provides a chronological listing according t o Early
Start time of each activity, thus offering a form of
''tickler file" by date. Late Start or any other
computed date may be used instead of ES.
Responsibility Code Each activity mav be assigned a code representing
the person, apncy or subcontractor responsible for
it. A sort on this code groups the activities by such
responsibilities, providing a convenient report for
subcontractors and others.
Designated Key Events or Activities Provides a summary report for too management.

of the project is now day 48 instead of the desired day 40. The total float has
gone from the initial value of zero to -8, or 8 days behind schedule.
The float sort is not the only useful sequence of the output. For large net-
works an I-J sort is useful for quickly locating data on particular activities. A
sort by one of the date columns can also be of value. For example, a sort by the
L F column provides a list in chronological order, with the activities that should
be finished first listed at the top. All of these sort options and others are usually
available through simple sort key codes that can be specified by the user. Most
users will specify two or three different sorts for each update report. A sum-
mary of some of the more useful sorts is given in Table 11-1.

PRECEDENCE METHOD

As mentioned earlier, this chapter employs only the arrow scheme to illustrate
computer input and output. For an example of computer reports under prece-
dence diagramming, see Chapter 4, Figures 4-13,4-14,4-17, and 4-18.

HOW TO FIND COMPUTER SERVICES

The availability of computer semices is a twofold question; first, is a computer

available, and second, is an adequate critical path program available for that
computer? Organizations not possessing a computer may survey their city or
348 11 I Advancad Topics

nearby cities for computer installations that rent time on the equipment. The
local representatives of the computer manufacturers are good sources for this in-
formation. Rental time is offered by many banks and industrial firms as well as
computer service companies, and such services are often found in relatively small
cities and towns. With the recent advances in mini- and microcomputers it has
become possible to purchase one with critical path software for only two or
three thousand dollars.
Having determined what equipment is available, one should determine next
whether a suitable critical path program is available for that equipment. The or-
ganization operating the computer may have a suitable program in its library, the
manufacturer may be able to supply one, or it may be necessary to seek a pro-
gram from other sources. The appendix to this chapter lists a number of CPM
and PERT programs available t o the public. These programs differ significantly
in a variety of ways, however, and the differences are worthy of close attention
by the potential user.
Computer software companies are also good sources to investigate, since a
number of them offer CPM software packages for sale. Purchasing a package is
worthwhile if (a) it is expected that the package will be used often, (b) the pack-
age makes it possible to utilize the user's in-house computer, and (c) the package
has valuable features not otherwise available.
It is also possible t o obtain network computational service on a time-shared
basis. The user provides the input through a rented or purchased terminal. The
input is transmitted over telephone lines to a large computer at some other loca-
tion, perhaps in another city, or via satellite to another country. The computer
processes the input immediately, while it is simultaneously processing a number
of other jobs for other users. The critical path output comes directly back to
the user's terminal.
In any case, it is important to obtain and study the user's manunl that is pub-
lished for each available program. The manual describes the characteristics of
the software and how to provide input for it. When attempting to use software
for the first time, it is a good idea to test the published procedures with a small
network such as the one in Figure 11-3. In this way one can verify the results by
hand computation and also check the logic on updating, calendar dating, etc.

COMPARATIVE FEATURES

In order to summarize the advantages and disadvantages of various critical path

programs (the reader is reminded that the term critical path progrums or soft-
ware as used in this chapter refers to programs for CPM, PERT, and similar net-
work methods), certain key features are listed below and commented upon.
The features discussed are by no means exhaustive of the points on which
critical path programs differ. They are among the most significant, however, to
the average potential user.
 Computer Processing 349

1 . Arrow vs. Node Format. It is necessary that the network and the soft-
ware be consistent in the use of either the arrow or the node network format.
Thus, the available software might dictate which network format is to be used.
An increasing number of software packages now will accommodate either the
arrow or node format, or the more advanced precedence scheme.
If the user has already drawn a network in the arrow format and then finds
that the only available software requires the node format, the network can be
redrawn into the node format. Conversion from arrow to node is relatively easy.
Conversion from node to arrow, however, may not be so easy unless the user is
familiar with the pitfalls of arrow networking, such as the creation of false de-
pendencies. (See "Common Pitfalls," Chapter 2.)
2. Event Numbering. A few of the oldest CPM programs require that events
be numbered in ascending order, that is, each activity's successor event number
must be larger than its predecessor event number. This is a severe restriction, for
it inhibits the flexibility of the network and causes event-number bookkeeping
problems. Most programs now permit random numbering of events (or activities
in the node format), provided that no two nodes or activities receive identical
numbers.
3. Capacity. The capacity of critical path programs is usually expressed in
terms of the number of activities permitted. Capacities vary from a few hundred
to at least 500,000 activities.
If a network exceeds the capacity of the program, it is not usually practical t o
divide the network into parts for separate computation. The interaction of for-
ward and backward passes along all paths can make such subdivision highly com-
plex. A better approach is to condense the network, as explained in Chapter 3.
(See also Network Condensation, feature No. 12 below.)
4 . Calendar Dates. Many programs have a calendar dating option, which will
provide all output dates in the fonn of 02/03/85 or 03FEB85. To use this op-
tion the user needs only t o input the base starting date for the fust event in the
project. A few other options are sometimes available, such as whether the calen-
dar computation is to be based on 5-, 6-, or 7-day weeks, and whether there are
holidays. The calendar dating option is highly desirable, although some indus-
trial projects are scheduled by hours and would not use calendar dates. Some
programs assume that start dates refer t o the beginning of the day given, whereas
fmish dates are assumed to be at the end of the day given. This convention is
logical and presents no problem as long as all the users are aware of it.
5. Scheduled Dates. Most programs will accept scheduled dates assigned t o
the terminal events in the network, and backward passes are made from these
scheduled dates, rather than merely "turning around" on the terminal event's
earliest expected date as computed in the forward pass. The acceptance of
scheduled dates means that the slack figures will be related to the scheduled
dates, and the critical path may have positive, zero, or negative float. (The criti-
cal path is defined as the path of least float.) Without this feature, one must per-
350 11 I Advanced Topics

form hand or mental computations to determine the relationship of the ex-

pected completion date and the scheduled date.
In some programs scheduled dates are permitted at any intermediate event. In
such programs the float computation is based on either the scheduled date or the
computed latest allowed date, whichever is more constraining (earliest). This
computation can result in a discontinuous critical path, i.e., one that may begin
in the middle of a network.
Projects may indeed have more than one starting point and/or more than one
objective, as discussed in Chapter 4. Where this is the case, the computations are
inconvenient and sometimes incorrect if all the "dangling" events are artificially
brought together t o single initial and terminal events, which is a patch-up proce-
dure sometimes used.
6 . Multiple Initial and Terminal Events. Some critical path programs require
that networks have only one initial event and only one terminal event. Other
programs permit multiple initial and terminal events, which is advantageous for
the situations described above. Also, if it is desired to merge two or more par-
allel projects for a single critical path computation, the multiple initial and ter-
minal event capability greatly simplifies the mechanics of the procedure.
7 . Error Detection. Critical path programs vary widely in their capability t o
detect and diagnose network errors and inconsistencies such as loops, nonunique
activities, improper time estimates, and excessive terminaI events. Most pro-
grams will not only detect loops and stop, but will print out the event numbers
in the loop. This greatly aids in the manual search for the network or input
error, which is usually two different events with the same number. Other types
of error are less difficult to detect by programmed routines, and thus most pro-
grams wiIl provide adequate detection and diagnosis.
8 . Output Sorts. The way in which critical path output data are sequenced,
or sorted, affects the utility of the report as a management tool. A list of the
most useful sorts and their applications is given in Table 11-1. Most software
offers all of these options and others.
Some programs also offer options to sort on major and minor keys, such as
"ES within float" (or float major, ES minor key).
9. Report Generator. Although most programs provide their output in a
certain fmed format, some programs permit a high degree of flexibility in for-
mat. By means of "report generator" routines, the user may select which infor-
mation (from a list of 20 t o 3 0 available fields of data) is desired and may spec-
ify the columnar sequence of the fields. Thus, the user can tailor the reports to
particular needs, eliminating aU undesired data. (See Figures 11 -9 and 11-10.)
10. Graphical Output. In order to provide readable summaries for manage-
ment, it is often desirable to convert tabular data into graphical formats. Graph-
ical outputs of various kinds have become increasingly popular features of criti-
cal path software. These include bar charts, resource requirement distributions,
PROJECT 072
ACT1 ; C T I V I l Y
V I T Y CE.SRIPTION
TWO STORY S T E E L I C O N C R E T E
RESOURCE
NAME
CLASSICAL C R I T I C A L
REVD R E 5
ESTIMATE
TIME
P A T H REPORT TARGET D A T E 09/28/82
PROJECTED P R O J E C T E D L A T E S T A R L A T E
R E M A I N I N G START FINISH DATE FINISH FLOAT
PAGE 1
.=
b
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
STEEL FOREMEN
LABOR FOREMEN
H E A T I N G FOREMEN
P L U M B I N G FOREMEN
LABOR FOREMEN
LABOR FOREMEN
H E A T I N G FOREMEN
P L U N B I N G FOREMEN
LABOR FOREMEN
MASONRY FOREMEN
MASONRY FOREMEN
LABOR FOREMEN
E L E C T I C A L FOREMEN
H E A T I N G FOREMEN
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
ADMINISTRATION
P L U m B I N G FOREMEN
ADMINISTRATION
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
LABOR FOREMEN
E L E C T I C A L FORENEN
E L E C T I C A L FOREMEN
E L E C T I C A L FOREMEN
H E A T I N G FOREMEN
STEEL FOREMEN
'"Q FOREMEN

Figure 11-98 Result of input specification shown in Figure 11-9A is a report with critical path [zero float in thiscase) activities listed first.
llluEtrations of PAC llTM Report Writer courtesy of International Systems, Inc., King of Prussia, Pennsylvania.
3 .WJOJ 40 pua i e ino
-Ael leurunlo3 aioN .aldluexa s!qa u! paaenrnll! s! a h y x d aJeMyos Jai!JMFlOdaJ e 40 suealu A q u e q u~ u e e~ 40 uo!ie3!4!3adS WOL-LL wnB!j
P R I N T SHOWING ' PHASE' ' I C T I V I T I ' 'RESOURCE' 'PROJ F I N I S H DATE'

-
<n --
'REVD F I N I S H DATE' 'CALENDAR'
-- --
-..l T. F n ' P. F.n . I E C T. '
S EL E C T IF RECORD^ = .'7".
EJECT PACE ON 'PROJECT'.
T I T L E SIiOWING "PROJECT DESCRIPTION" 'PROJECT DESC' "PROJECT" ' P f O l E C T ' .
GAtlTT CHART FROM " 1 0 0 1 8 2 " TO " 0 3 3 1 8 3 " SHOWING 'PROJ F I N I S H D L T E
R E L A T I V E 1 0 'REVD F I N I S H DATE' U S I N G I I T F R A L S "+-" " '-
"OCT NOV DEC JAN FEB
"1 1 1 1 I 1 1

PROJECT D E S C R I P T I O N VERY IMPORTANT PROJECT PROJECT PROJ 3 0 0 DATE 0 4 / 2 2 / 8 1 PAGE 3

ACT1 RE5 PROJECTED REVISED OCT NOV DEC JAN FEB MAR kPR
PH VlTY ID FINISH FINISH 1 1 1 1 1 1 1

1 0010 JAN 10/03/82 10/08/82 -

1 0020 DOT 11/15/12 11/23/12 +++
1 0030 ART 10/15/82 10/15/82

1 0040 KEN 11/15/82 12/01/12 +++++

2 0050 BEN 12/07/82 12/06/12

2 0060 001 11/01182 12/08182 +++*+*I++

2 OD70 ART 02/10/83 12/30/82 .......--.---

2 0080 DOT 12/14/82 12/16/82 +*

2 0090 JAN 03/04/63 01121181 - - - - - - - - - - -- -
2 0090 KEN 02/17/83 01/07/83 .....-. - - -- - -
.--..--------
2 0100 PAT 03/18/83 02/04/83

Figure 11-108 Gantt chart resulting from entry in Figure 11-IDA shows first the edited in?; instructions then the chart. Note that plus
signs denote activities with positive float, minus signs for negative float. illustrations of PAC I I system courtesy International Systems, Inc.,
King of Prussia, Pennsylvania.
 Computer Procesing 355

and even the network diagrams themselves. The charting devices used include
the standard line printer on most computers, the CRT display screen, and con-
tinuous line plotters. For illustrations of a few of the graphic outputs available,
see Figures 11-10 through 11-15.
11. Updating. A good software package will provide for updating by excep-
tion. The original network is stored on disk or tape, and progress reports are in-
put to update and revise the network and schedule projections. The input trans-
actions should consist only of changes in the network and actual progress, such
as the activity start and completion dates since the last update as illustrated ear-
lier in this chapter. The entire network need not be input again.
However, there are significant differences in how the updating function is
handled among the available programs. For example, some programs permit
only the input of finish dates of activities; the start dates are ignored. This
means that if an activity has started late and is not yet finished, the late condi-
tion will not be recognized by the program, and the output will be incorrect to
that extent. Also, if only finish dates are accepted, it is not possible to compute
and record the actual duration of each activity for comparison with the esti-
mated duration.
A similar but more serious error is to assume that, on a path having had some
progress dates reported, the next unreported event date on the path will occur at
its earliest time (ES or EF), regardless of the effective date of the input. To
understand the effect of this assumption, review the market survey network and
the sample progress report in Figures 11-6 and 11-8. Activity 2-3 is pertinent,
for it has not started as of the reporting date, day 12. The correct handling of
this condition is to compare the ES for the activity with the reporting date, and
take the larger of the two as the basis for beginning the forward pass. If the re-
port date is larger, one day may be added. Thus, since 2-3 had not started by
the end of day 12, the earliest that it could now start is day 13. Therefore, the
correct ES for 2-3 is 13, the E F becomes 18, and the project completion date is
extended.
By the incorrect procedure, the ES for 2-3 would be assumed to be 2 (which
is impossible since day 2 is now ten days ago), the EF would become 7, and this
path of the project would apparently be a day ahead of schedule. Incorrect re-
sults of this type have caused misunderstanding and confusion for many users.
Programs using this incorrect procedure should be avoided.
12. Network Condensation. Routines have been developed which in effect
condense large networks into smaller ones. One such procedure involves three
phases: (a) the condensation of large, detailed networks into smaller ones, (b)
the integration of two or more condensed networks, and (c) the expansion of
the condensed and integrated networks back into large, separate, detailed net-
works. The condensation phase involves the determination of the longest path
between each sequential pair of preselected key events in the detailed network.
 . a a s n q a m a W ' a 6 p ! l q L L I ~'.out 'auawdolanea g ale%-
a 3 a f o r d A s w n q 'waasAs ramduo3 Z/U~~IJI&~ aq8 A q paanpord w e q a J B ~PUB 'sauna an!ze~n~una
'SWBIBOIS!~ 40 uo!gerwnlll ~ 1 . wnn!j
~ 1

YO1IVIN3YtlUlSYl L13A1130 019

- NOllV1113UnlllSYl 3SVH9Urld 009

- I N Y d l n 0 3 W31Y13313 11VISNl OCS

- 1N3Udln03 W31U13313 U3A1138 02%
lN3Ydltl03 W31ll13313 3SVH3Ynd 015

. I ~ . P ? ~ r. v w . w . .aw.p3?4r5..3:~.y
w . R ~ . ~ ? : . ? ? ~ . . B w.PA..&*"~
........ rtr ....... vjr.. . . . .r o .......
~ fir:::
,,....
,,,,,.......
ipr .........5.
25
7,,,,7,7,,,1WIIII :
0.0

,,,T~TI~ 1111 1 m a 3 1 3 1 a ~ ~ ~ ' ~ ~:

! .OIOE

! '00&*7

! 'OOoEr

'vmhuau aql u! '

~ml,(~aoaaqlEW!IPJ!~U! T
'sawn3 mol#qseoen!aelnruno :
a l n p q ~ r(1)are1 pue (3)A11e3 :'ooort
f '00001

xxxn ,f .ooosc
LW,XIII

seruna pue sUJI3160lS!~

 Computer Processing 357

PROJECT/2
G R A P H I C REPORT W R I T E R
RESOURCE 1 1 0 . 0 0 - LABORERS

WEEKLY RESOURCE USAGE CUMULAIIVE RESOURCE USAGE

MhNOAIS PERCENT

@ ~~!~"A~%REs(IuRcE
T R H
A G SC
LABORERS
ESI
,1000
.,
,**. kkiTARG
!R*:S SCH RESOURCE
LABORERS
1
1\000
E S cun

D
;fgW
; YESOURCE ! I 0 0 0 MANDAYS RESOURCE 1 1 0 0 0
r R E S scn 1 cun

Figure 11-12 Example of computerproduced graphic report of reswrce utilization, this

one showing both weekly and cumulative figures for the resource under two different proj-
ect schedules. Courtesy Project Software & Development, Inc., Cambridge, Massachusetts.

The result of this phase is a network consisting of the key events only, which are
connected by single activities instead of groups of activities.
In the integration phase the computer processes two or more condensed net-
works as though they were a single network, each condensed network having one
or more common events (interfaces) with another network. The results of this
process are new earliest and latest times for each key event. The expansion
phase, then, utilizes the new restraints on the interface events to compute new
schedules for all activities in the detailed networks. The output of the expansion
phase is in the same format as the output for the original detailed network, but
the earliest and latest times and slack figures reflect the new restraints imposed
on the interfaces.' Hence, this network condensation-integration-expansion pro-
cedure provides extraordinary capacity for processing the largest networks and
groups of networks. It also aids in the preparation of summarized reports for
management.
13. Statistical Anabsis. Originally the controversy of the three-time esti-
mate probabilistic approach versus the single-estimate deterministic approach to
critical path computations provided a clear distinction between PERT and CPM.
 358 11I Advanced Topics

mmi€as

300
BEFORE SCHEOULING
RESOURCE LOADING FOR CARPENTERS

m LOADING
MY
nm

, I
I

~ ~ w m r n i m n f f i ~ m m m ~ ~ ~ ~ n ~ ~ m ( f
-------.-------.------,981-------------.--2----.-.-----.----19Bp-------------->
PROJECT TIHESCALE (MONTHS)

Figure 11-13A Capabilities of graphics plotters are illustrated in this example of a resource
loading histogram for a project utilizing carpenters. Original report is i n four colors.

mmfE warm
AFTER SCHEOULING
RESOURCE LOADING FOR CARPENTERS
3m

270 - LOADlNG

- UNDERLOAD
240
OVERLOAD
ZlYI -

.--.
PROJECT TIMESCALE (MONTHS>

Figure 11-138 After the software has scheduled the project within the resource constraint,
the resulting histogram displays the extended project duration and the new monthly require-
ments for carpenters. Illustrations of ARTEMIS system courtesy of Metier Management
Systems, Inc., Houston, Texas.
 Computer Processing 359

While the distinction still exists among many critical path specialists and is main-
tained in this text, the two terms have been used loosely in practice and are no
longer mutually exclusive. For example, some programs permit three time esti-
mates in the input and obtain the usual weighted average of these figures, but do
not make any summations of variance or computations of probability. Another
system used by a govenunent agency is labeled 'TERT" but makes no use of
three time estimates or statistics. Consequently, if one desires the probabilistic
approach he must be careful not to rely on the title "PERYor the three-estimate
input, but must determine exactly what probabilities are computed by the pro-
grams in question. The most common probability computation is an approxima-
tion to the probability of meeting a scheduled date, usually the scheduled date
for the completion of the project.
Those who are interested in the CPM (deterministic) approach but who find
only PERT (probabilistic) programs available to them do not have a serious ob-
stacle. Almost any probabilistic program may be used as a deterministic one by
simply entering the single (expected) time estimate as each of the three estimates.
14. Interactive Processing. A trend in software is the facility for the user to
interact with the program while processing is underway. Using keyboard data
entry, for example, the program may make error checks on each set of activity
data, and signal the user immediately when errors are found. After all the data
are entered, initial computations may be performed and displayed so that the
user can change the network in order to obtain a more acceptable schedule be-
fore printing out the reports. Frequent users of network methods find interac-
tive processing advantageous.

TRENDS I N SOFTWARE

In the first twenty years of critical path network technology the computer pro-
grams for network processing have undergone evolutionary change as well as pro-
liferation. The growth in the market for commercial software and the features
that have proven popular and unpopular provide a measure of the practical appli-
cations of network methods over the two decades. Trends for the future can
also be discerned. (Material for this section is based on experience of the authors
with several commercial programs, contacts with other users and software ven-
dors, and several excellent surveys of software published between 1964 and
1980. (See References 2-6.)
The earliest software often had several inconvenient limitations. The pro-
grams were almost all designed solely for the arrow networking scheme. Many
required I < J node numbers for each activity, provided only one output tabula-
tion, did not provide calendar dates, and were restricted to a single network with
only one starting and one ending node. In the early sixties resource allocation
algorithms were just being considered by graduate students and others and were
not yet available in commercial programs.
360 11 I Advanwd T o p i a

SAIPLI ) I C I I V I I I : 000000000000 S = l I L R T 10 $ 1 1 1 1 LLO IllMI IN GllS 1

0 "1111 0 c = r I l l $ n 1 0 s r l n l LA6
0 #a#I 0 I : F I I I S U 1 0 T I H I S R LAO
RIB At1 116
O=DUR r;rru
tbRL1 S I I l l I
t l l l l l FlWIEU
OOOOOOOO~OOO ~=C~ITICL LCTIUIII I=COIIPLII~D (ICTIUITI

111111 0 1 1 1 I S IllPRl7 UORliH6 8 . YPSIO).I O I IOU TU1 Y1B TMU FRl

Figure 11-14 The ability of computer systems t o "draw" networks is displayed in this por-
tion of a precedence network plotted on paper by a PAC graphics printer. Courtesy Interna-
tional Systems, Inc.. King of Prussia. Pennsylvania.
 Computer Processing 361

Figure 11-14 Continued.

362 11 I Advanced Topics

Figure 11-15 Example of precedence network displayed on video terminal by OWIKNET

software, part of PROJECT12 system. With hand-held cursor the user may point to activi-
ties and indicate paths between them; the Wstem then draws the precedence lines and
prompts the user for lag relationships. Courtesy Project Software L Development, Inc.,
Cambridge. Massachusetts.

On the other hand, some of the earliest programs contained rather sophisti-
cated mathematical features that have since almost disappeared from commercial
software. The statistical and probability features originally associated with
PERT have been the most notable casualties, due to the lack of user interest.
Most programs now operate with a single time estimate per activity.
Similarly, users have not maintained interest in the powerful timecost trade-
off procedure originated by Kelley and Walker in the first CriticalPath Method
computer program. While managers still acknowledge the validity and irnpor-
tance of the timelcost optimization concept (particularly with the dramatic in-
flation in construction and other costs in the late seventies), they tend to deal
with the problem in other ways that do not require detailed time-cost analysis of
each activity in a network. Another reason may be that resource allocation
problems may be more constraining. Why expend a great effort to find an opti-
mal time-cost trade-off if the resultant schedule is infeasible due to resource limi-
tations? As yet, no software has been able to optimize resources as well as time-
cost trade-offs.
Other interesting trends in the programs, indicating user needs and applica-
tions as well as advances in computer hardware and software, are categorized and
commented upon below.
 Computer Processing 363

1. Network Scheme. The shift toward node or precedence diagramming has

continued, judging from the software available.
2. Graphic Reports. Report formats have tended strongly toward such
graphic displays as bar charts, resource histograms, and even network plots.
These forms of output indicate emphasis upon extensive analysis of resource
allocation problems, more use of network methods throughout the course of
projects (using updating and progress reporting features such as revised network
plots), and displays for clients or upper levels of management (who are informed
and impressed more with graphs than with tables of data).
3. Resource Allocation. Although absent from the earliest programs, vari-
ous kinds of resource allocation algorithms have now become common in the
commercially available packages. These include time-constrained and resource-
constrained procedures as well as a few optimizing routines. Popularity of these
procedures indicates the value of network methods in analyzing an important
practical problem in a wide range of project types-the utilization of limited
resources.
4 . Hardware/Software Advances. While the early programs required medium-
to large-scale computers, the power of those machines is now available in mini-
and microcomputers. Network software is now available for these very small
configurations. Some programs also operate on an interactive basis with the
user, through a keyboard or CRT terminal. Some programs are available through
computer time-sharing services.
5. Vendors. Through the sixties most of the network software packages
were available through the major computer manufacturers. The predominant
trend in the seventies was toward the development of improved packages for sale
or lease by software firms, some of which specialized in network services and
software. Time-sharing services have also entered the business of critical path
processing, which appears to have developed into a significant segment of the
burgeoning computer software market in the U.S.

REPRESENTATIVE SOFTWARE

To illustrate the variety and sources of software available in 1981, a small but
representative list of programs is provided in Table 11-2. The list was compiled
primarily from a comprehensive survey published by the Project Management
Institute, plus other sources. The listing for each program was verified by the
vendor. The vendors are listed at the end of this chapter.
All of the programs listed are available t o the public in some manner through
purchase, lease, or user charges. Not listed are any of the scores (or perhaps hun-
dreds) of proprietary programs used internally by corporations, consultants, or
universities.
The table does not attempt to list all of the special features offered, but is
364 11 I Advanced Topics

limited to a few of the more popular and significant features. All of the pro-
grams are presumed to incorporate the basic network scheduling calculations de-
scribed in Chapter 4 and earlier in this chapter, including earliest and latest start
and finish times for each activity, and total float. For more details on each pro-
gram, including price information, contact the vendor.
The reader is cautioned not to infer any evaluations of quality or cost-
effectiveness among the software packages listed in Table 11-2. It is possible
that a program with comparatively few features may be among the most effi-
cient, reliable, and cost-effective in certain applications. The authors disclaim
any comparative evaluations of the programs listed, nor do we suggest that these
are the best available at the time of this publication. For more up-to-date and
qualitative information of this kind, review current literature and contact the
Project Management Institute, P.O. Box 43, Drexel Hill, Pennsylvania 19026.

SUMMARY

The question of when and how t o use computer programs for network computa-
tion has been covered in this chapter, along with examples of input procedures
and output reports. It is stressed that the software packages are readily available
to the public and that they are not difficult to learn to use, even for persons
totally unfamiliar with data processing. There are some significant differences in
the features available in the programs, and the,user should study these carefully
before making a choice.
There is also a wide variety of software packages available for purchase or
lease. The trends in the development of this software reveal the popularity of
network methods in practical applications. A representative list of software
commercially available is provided.

SOFTWARE VENDOR ADDRESSES

Further details about features, prices, installation services and other information
about the software packages in Table 11-2 should be addressed directly to the
appropriate vendor as listed below. The list was prepared and verified in Janu-
ary, 1981. Addresses and contact names, of course, are subject to change after
that date. The authors do not assume responsibility for further information on
the packages or on vendor addresses.
Program Name Vendor Contact Person
APECS ADP Network Services, Inc. Mr. Richard W. Rogers,
180 Jackson Plaza Manager, Project
Ann Arbor, MI 48106 Management Applications
 TaMe 11-2. Representative Software

C
APECS
PROGRAM NAME SOURCE LANGUAGE AND
HARDWARE REQUIRED

MCRO 10, FORTRAN

Available o n l y through
ADP Network Services
NEMORK SCHEME
AND SIZE

Rrrow o r Precedence
64,000 a c t i v i t i e s
Y

X X
S ' J U . ?

X X X X X X X F u l l y i n t e r a c t i v e . Integrated
cost/schedule analysis. Many graphic
output options.
Hewlett Packard lODO Armw o r Precedence X X X K X X X X X " User- friendly' harhrare/soft*are system
mini w i t h 20 Mbyte disc 32,000 a c t i v i t i e s w i t h f l e x i b l e p r i n t e r o r graphic
drive reporting.

cm FORTRAN Available on several timeshare services.

50K. d i s k

CPMIS COBOL Armw X X X X X X X N e t w r k s i z e a f u n c t i o n of cmputer

IBM 360130 w i t h 250K 30,000 activities memory. Unlimited nunber o f projects
core, 4 tapes. 22314's i n a dataset. System i s modifyable t o
user needs.
KRONOS ANS COBOL
65K
NodeIPrecedence
1,000 a c t i v i t i e s
Arrmr
IXxx Capacity i s expandable. Sumdry bar
charting, other r e p o r t i n g features.
I n t e r a c t i v e , menu-driven, m n y graphic
440 a c t i v i t i e s and r e p o r t output options. Resource and
cost plots.
 Table 11-2 Continued

18
A
.
Z P B o U
. L,

YO s - z , u u
rnD 0 . 2 Y C Y ) O
O

SWRCE LANGUAGE AND NEWORK SCHEME

,,zrz x,eSz
.-
Y t l L Y l

% U r n %
HARMARE REQUIRED AND SIZE * c1Ju.-
C ~
8
; ~ : ~ Y
0
~ Y *

S"P L , ' < S l t f .

P I Y I > D a . - W Y
~ U L " ~ ~ U ' c C C L Z

FORTRAN
From m i n i ' s t o l a r g e
scale computers

COBOL FORTRAN
~ v al i b l e through General
I 200-10,000 a c t i v i t i e s
X

X
X

X
X

X
X

X
X

X
X

X
X

X
X Available i n 5 model sizes f o r mll to
l a r g e canputerr and m r k programs.

Cost curve reports. Resource forecasti

Unlimited n-r o f resources per
E l e c t r i c Mark I 1 1 Systen activity.

WCS (Management COBOL, FORTRAN, BPI Arm o r Precedence X X X X X X X X Resource constrained scheduling.
Scheduling and Control IBn 3601370 42,600 a c t i v i t i e s reporting, s e l e c t i o n c r i t e r i a . Graphics
System) OS, MYT interface.

FORTRAN. COBOL. Arrow o r Precedence X X X X X X X X Data base a v a i l a b l e f o r user-defined

Assembler 11,000 a c t i v i t i e s for reports.
1100 CPU time processing x 4095
networks

FORTRAN and Assembler Arro* o r Precedence X X X X X X X X An integrated database Ranayanent system

Medium t o l a r g e scale Unlimited f o r schedule and cost control.

PAC 11 ANS COBOL Node o r Arrow X X X X X X X X X S k i l l s scheduling by resource. Effec-

Any computer w i t h COBOL tiveness scheduling.
c m p i l e r and minimun Basically unl~mited I n multi-project m d e -
core, tape or disk activities develops c r i t i c a l path for each p r o j e c t
i n c l u d i n g t o t a l f l o a t for each project.
i n t e r a c t i v e f a c i l i t i e s , and
Report Writer.

I PCM (Project Cost

Model )
COBOL or FORTRAN
Wort small o r l a r g e
118achines
Arrow o r Precedence
10.000 a c t i v i t i e s
X X X X X X X Resource optimization. F u l l cost and
cash flow analysis and forecasting.
 SWRCE LAHGUAGE AND NETWORK SCHEME
HARDWARE REQUIRED AND SIZE

WCS (Project
Management and Control
System)
FORTRAN
Large Scale

Assembler
Any I?360/370, 303X,
I Arrow o r Precedence
Unlimited

A r r W o r Precedence
8 million activities
X X X X X X X

X X X X X X X X
Resource Leveling
I n t e r a c t i v e I n p u t and Reporting
Capabilities.

Extensive cost-performance analysis.

43XX w i t h supporting 05
type operating system

PROJACS
Assembler
1Bl 360 05 2561:
PL/1 and Assembler
I M 360/370. 303X, 43XX
I Arrow o r Precedence
64.000 a c t i v i t i e s
Arrow o r Precedence
32,000 a c t i v i t i e s
X X X X X X X

X Y X X X X X
X

X
Resource p l o t s ; repart c m a n d language.

Standard module (fragnet) l i b r a r y

capability.
PROJECT CONTROLJ70

PROJECT/Z
AHS COBOL
250K disk, 150K core
I Precedence
Unlirni ted
A r r r o r Precedence
X X X X X X X

X X X X X X X X
Report generator.

English language canands, r e p o r t writer.

I
Any IBM OSIYS, any UWIVAC 32.000 a c t i v i t i e s resource plots, cost curves, bar c h a r t
1100 under EXEC 8; UNIVA p l o t s , create networks on graphic
Series 90 under YHOS; any
D i g i t a l Equipnent YAX
terminal -QUIKNET.
under VMS

on Mini Conputern)
1
PRDHIWl (Project Contro FORTRAN
64K words
I Arrow o r Precedence
32,000 a c t i v i t i e s
I n t e r a c t i v e Input, r e p o r t w r i t e r .
resource plots.
I
 Computer Proosging 389

Rogmm Name Vendor Contact Person

ARTEMIS Metier Management Systems, Mr. S. Wayne Wyatt
Inc.
10175 Hamin Drive, #I00
Houston, TX 77036
Glenn L.White Company Mr. Glenn L. White
10560 Main Street, #406
Fairfax, VA 22050
KRONOS American Software, Inc. Dr. Thomas L. Newberry
343 East Paces Ferry Road
Atlanta, GA 30305
Sheppard Software Company Mr. Leland C. Sheppard
1523 Coronach Avenue
Sunnyvale, CA 94087
MISTER Shirley Software Systems Mr. Walter W. Shirley
1936 Huntington Drive
South Pasadena, CA 91030
MPM (Multi-Project Florida Power Corporation E. L. Schons
Management) P.O. Box 14042
St. Petersburg, FL 33733
MSCS (Management McDonneU Douglas Mr. Oscar H. Stepanek
Scheduling and Automation Company
Control System) P.O. Box 516
St. Louis, MO 63166
OPTIMA 1100 Sperry Univac J. D. Heitner
P.O. Box 500
Blue BeU, PA 19424
OSCAR On-Lie Systems, Inc. E. R. Artus
115 Evergreen Heights Drive
Pittsburgh, PA 15229
PAC II International Systems, Inc. Mr. Joseph S. Herbets
890 Valley Forge Plaza
King of Prussia, PA 19406
PCM (Project Cost Project Software Ltd. Mr. Martin Barnes
Model) 23 CoUege HiU
London EC4R 2RT, England
PMCS (Roject Management Honeywell Information Mr. G. E. Hanson
and Control System) Systems, Inc.
Large Information Systems
Div.
P.O. Box 6000
Phoenix,AZ 85005
370 11 I Advanced Topics

Program Name Vendor Co~tactPerson

PMS lv Scientific Marketing Mr. Peter V. Norden
IBM Corporation
1133 Westchester Avenue
White Plains, NY 10604
PROJACS Scientific Marketing Mr. Peter V. Norden
IBM Corporation
1133 Westchester Avenue
White Plains, NY 10604
Project Software and Mr. Ray Haarstick,
Development, Inc. Vice President, Sales
14 Story Street
Cambridge, MA 02138
Atlantic Software, Inc. Mr. Ted Stein, Manager
320 Walnut Street Network Sales
Philadelphi, PA 19106
PROMINI (Project Control K&.HComputer Systems, Inc. MI. Sam Phelan
on MiniComputers) 48 Woodport Road
P.O. Box 4
Sparta, NJ 07871
PROSE Construction Industry U. Komgold or
Computer Consultants Ltd. R. F. Gurr
(CINCOM)
620 Dorchester Blvd. West
Montreal, Quebec
Canada H3B-IN8
Hollander Associates Mr. G. L. Hollander
P.O. Box 2276
Fullerton, CA 92633

REFERENCES

1. Prostick, J. M., "Network Integration, a Tool for Better Management Plan-

ning," Presented at Meeting of Operations Research Society of America,
Philadelphia, November 7-9, 1962.
2. Petersen, Perry, "Project Control Systems," Datamation, June 1979.
3. Phillips, Cecil R., "Fifteen Key Features of Computer Programs for CPM and
PERT," Journal of Industrial Engineering, January-February 1964.
4. Smith, Larry A,, and Peter Mahler, "Comparing Commercially Available
CPM/PERT Computer Programs," Industrial Engineering, April 1978.
5. West, 0.E. e t al, Survey of CPM Scheduling Software Packages and Related
Project Control Programs, Project Management Institute, Drexel Hill, Penn-
sylvania, January 1980.
 Computer Prneersing 371

6. Woolpert, Bruce, "Computer Software for Project Planning and Control," un-
published student paper, Stanford University, Graduate School of Business,
May 1976.

EXERCISES

1. Using the network of the survey project in Figure 11-3, update the project
schedule with the following progress as of the end of day 8:
Actual Actual
Activity Start Date Finish Date Change
1-2 1 3
2- 4 3 Change time estimate to 8 days.
2-3 5 No change in time estimate.
4-6 Change time estimate to 8 days.

Assume that the project's scheduled completion day is day 37. Manually
prepare an output report similar t o the one in Figure 11-8 and answer the fol-
lowing questions:

a. Is the project on schedule?

b. What is the critical path?
c. List the features of a computer program that would be necessary t o
make the same updating calculations and produce the same report you
just did by hand.

Using the network in Figure 11-3, assume that the same questionnaire is t o be
used in another survey project. The deadline for having the printed question-
naire on hand t o conduct the other survey is day 15. Thus, day 15 is a sched-
uled completion date for activity 4-6. With this added constraint, manually
compute the initial schedule for the project, producing a report similar t o
Figure 11-5. What features of a computer program would be needed t o make
the same calculations? Are these features available in all CPM software?
3. Redraw the network in ~ i ~ u11-3r e using the activity-on-node scheme. Com-
pute the initial schedule manually. What differences are there in the results
shown in Figure 11-5? If you used a computer program, how would the in-
put format differ from the one shown in Figure 11-4?
4. Repeat Exercise 1 using a CPM software package available t o you. Discuss
any differences in the output results.
5. Consider a research and development project for a new plastics product in-
volving over 2000 activities, a budget of $1 to $1.3 million, and an estimated
time frame of 20-26 months. Management wishes t o use network analysis t o
plan and monitor progress on the project. An updated network schedule is
desired the f i s t of every month, including graphic presentations of (a) the
372 11 I Advanced Topics

critical path activities and their schedule status, (b) summaries of costs t o
date compared with budget, and (c) listings of activities under the responsibil-
ity of each internal department and each outside vendor and contractor. If
you are t o prepare the project network and present the desired monthly re-
ports: Would you want t o use a computer program or make the analyses
manually? If you decide t o shop for computer software, what features would
you seek? Would any of the programs listed in Table 11-2 not meet your
needs?
6. Suppose that a particular software package offers the features of multiple ini-
tial and terminal events and the option of assigning different scheduled dates
t o the initial and terminal events. The software literature does not mention a
"multi-network" feature. Could the package be used to process multiple net-
works that are interrelated? If so, construct an example of three projects
that are interrelated but have different scheduled completion dates. Explain
how the package could be used t o properly compute the schedules of each
project and all the activities involved.
7. Consider Case 3 on page 339. Suppose the contractor wanted to use a net-
work for schedule updating and control throughout the project. What would
be some of his options for manual or computer processing?
8. Redraw the network in Figure 11-3 using precedence diagram logic including
the following special interdependencies:
Predecessor Successor
Activity Activity Relationship
4-5 Start-Start (Lag 5)
Finish-Finish (Lag 2)
7-8 Start-Start (Lag 51
Finish-Finish ( l a g 3)

Compute the initial schedule manually. What differences are there in the re-
sults shown in Figure I 1-5? Comment on the relative advantages of using a
computer t o process this network compared t o problem 3.
 SOLUTIONS TO EXERCISES

Chapter 2
1. Schedule inspector

Inspect machine

5. These are resource dependencies, indicating that there is only one crew of
carpenters for the forming work. The dashed arrows from events 32, 37, and
42 also show that some subsequent activities are dependent on the use of
forms that are removed in activities 31-32, 36-37, and 41-42. (The subse-
quent activities are not shown.) In this case the resource is the reusable
forms.
374 11 I Advancad Topics

Chapter 3
 Solutions to Exercises 375

5. To the extent practical, plan activities in parallel rather than in series. (Often
this means providing additional resources to avoid resource dependencies.)
Anticipate long lead-times, such as fabrication of customized equipment, and
plan for them t o begin as early in the project as practical.
7. Inflation of time estimates tends to force those activities to become critical,
thus increasing management attention and pressure.
9. The superintendent for a high-rise building project may want enough detail to
work out the most efficient sequence for a floor, because the time saved
would be accumulated over all floors. The president of a large firm is not
likely to want much detail, but rather just the schedules for a few key events.
A programming subcontractor would be especially interested in the true de-
pendency relationships between programming activities and all other activi-
ties in the project.

Chapter 4
1. a. No.
b. 2.
c. As late as possible, i.e., 6-7.
3. a. Critical Path 210-106-109 Slack - 2
b. No effect for a time of 16; however, a scheduled time of 12 for event
106 would make LIo6 = 12, and thus the path 210-106 would have a
new slack value of - 4.
5. c. Latest allowable start time for the deactivation of the lines is 205 hours.
d. Activities on the critical path are A-B-F-G-M-D-0.
7. a. Project is one day behind schedule. Path 3-4-5-8has a slack of -1.
9. a. and b.

5
- (0-11
2
2

7
2
+

7
17-21
4
6

11
6

11
(2-51
1
7

12

0
Dummy

0 °
0

0
- - - -
0

0
(0-3)
2
2

2
2

3
(3-41
6
7

8
,7

8
(4-51
4
11

12
11

12
(5-81
3
14

15\l5
Dummy
l5

-
finish
0 1 2 10 10 15 :15 O 15-
10-61 13-71 17-81
1 8 5
6 7 2 10 '10 15 Key:

1 4 (Activity no.)
(6-71 duration
3
7 10
376 11 I Advanced Topics

11. a.,b. No Splitting Allowed Splitting Allowed COmpuration

Values of
Activity ES EF LS LF F ES EF LS LF F aorp

A 0 10 0 10 0 0 1 0 0 1 0 0 -
8 3 11 8 @ 5 3 11 8 0 5 'p=8-4=4
C 0 20 7 27 7 0 2 0 7 2 7 7 -

D @ 8 7 13 5 @ 8 7 13 5 n=6-1=5
E 10 22 10 22 0 10 22 10 22 0 -

F @ 27 13 27 0 @f 27 13 27 0 ar=14-5=9
G @ @ 2 5 2 7 @ @ 0 2 5 2 7 @ -
Differences in times noted by circled values.

c. Critical paths are the same in both cases, i.e.

A-I0 (NC); FSO; E-12 (NC); FF5; F-14 (FC)

d. The allowance of splitting is not significant in this example. The critical

path is unchanged.

Chapter 5
1. Cumulative costs per period with the revised activity starts as indicated are as
follows:

Cumulative Cumulative
Period Cost Period cost
1 $ 500 9 $1 9800
2 1000 10 21 000
3 2500 11 21 900
4 4700 12 24000
5 6900 13 26100
6 10600 14 28200
7 13800 15 28900
8 17000

3. Overall t o date, the results are as follows:

Heater House: Behind Schedule and Over Budgeted Cost

Tanks: Behind Schedule and Over Budgeted Cost
Site Work: Behind Schedule and Over Budgeted Cost
 Solutions to Exercises 377

For most recent reporting period there are some bright spots:

Heater House: Behind Schedule and Over Cost

Tanks: Behind Schedule but Under Cost
Site Work: Ahead of Schedule and Under Cost

Based on total performance t o date, the project will experience a cost over-
run of $22,700 unless some remedial actions are taken (assuming this is
possible).

Chapter 7
1. c. The internal tangent approach appears appropriate for Resource A because
of the shape of the curve (relatively sharp increase in requirements be-
tween periods 3 and 13).
9 9 - I5
Lower bound estimate = -- 8.4 units/period
13- 3
The relatively gradual increase in requirements of Resource B over the
entire project duration makes the internal tangent approach inappropriate.
98
Average resource requirement = - = 5.4 units/period
18
d. Case 1
8.4
Criticality index, resource A = - = 1.05
8.0
5.4
Criticality index, resource B = - = 0.68
8.0
Case 2
8.4
Criticality index, resource A = - = 1.20
7.0
5.4
Criticality index, resource B = - = 0.77
7.0
e. Expected minimum project duration: determined by Resource A, i.e.,
142
Resource A: -= 17.75 1 1 8 days
8
3. The final schedule obtained using the SIO rule to establish the OSS is shown
below.
378 11 I Advancad Topics

Y q
B

- D
C
1
E 1
F
G
H

1 1 I I l l 1 1 1 1 1 1 1 1 1 1 1 / , , 1
0 3 5 89 13 15 19 21 24
Time

5. The final schedule is shown below.

fl
B
C
D
E
--
--
F
G
H
I
-J

I 1 I I 1 I 1 1 1 1 I I 1 I l l
3 5 8 1112 1618 21 24
Time

This sequence was obtained using the CRD rule to set the OSS and adding
both resources together. If only resource A is used (the critical resource), the
same solution is obtained.

Chapter 8
1. Project Indirect Total
Duration Direct Costs Co6D Costs Network Change6
12 610 1510 -
11 g10+40 =650 1470 DL 4
10 650 + 4 0 = 6 9 0 1430 D 13
9 6 9 0 + 8 0 =770 1470 A L 2, D T 4, G L 3
8 770+100=870 1530 F16.Gk2
7 8 7 0 + 1 3 0 = 1000 1620 B15,D13,F+5
3. Activity i-j.
Minimize (C1 - C2 ) 6 I - syb
 Solutions to Exercises 379

c,
Activity
direct
cons

-
slope = r

y v p i i
Yi,
dl dz d3
Activity Duration

Minimize (C2 - C 3 )fi2 - sY;

Subject to: d l + y ; + ( d 2 - d 3 ) ( 6 2- l ) + T i - Tid 0
62 = non-negative integer
-635 (6* - I )
(d2 - d 1 ) 6 3 Z Y >

Activity
'.I--& Slope = s
direct
costs
c,. - - -- -
I I
I
I
c,--+---+
I
I
----.
t+
I
Yt!,
I I
I
I
* y,,
d, d2 d3
Activity duration
380 11 I Advanced Topics

Minimize (Cz - C3) 6 2 - s l y ; -$ 2 ~ ;

Subject to: d l + y ; + ( d z - d 3 ) ( 6 z- 1 ) + $ + T i - Ti$ 0

6, = non-negative integer

ActiviN
direct
costs

I I
I I
I I
I I
I Slope = s,

I I
d4
'y,
dl d2 d3

Activity duration

Chapter 9
 Solutions t o Exercises 381

9. a. The activities that have two or more predecessor activities are 108, 11 3,
117, 119, 121, 125, 126, 127, 128, 129, 132, and 133. The predecessor
events of these activities are the ones where a merge event bias may occur.
Of these 12 cases, only activities 12 1, 125 and 126 will present merge
event bias problems. For the others, the expected means of the merging
paths are quite far apart, or else there is a great deal of correlation in the
two paths, so that the bias will again be negligible, e.g., activity 132 and
129.
It should also be noted that the merge event biases will occur on slack
paths, and hence will not affect the critical path, and hence the conven-
tional PERT analysis should be a good approximation in this example.
b. The critical path is made up of activities 102, 105, 108, 112, 115, 118,
120, 123, 124, 128, 13 1, 132 and 133. The sum of the estimated mean
times for this path is 382. The variance of this path can be approximated
by considering only activities 105 and 118. The sum of their variances is
1221.3 and the approximate standard deviation is 35.
c. P{T< 400) = P { Z < (400 - 382)/35} = 0.70.

Chapter 10
1. Some of the alternative networking systems (or alternative computer pack-
ages) t o consider include: 1. CPM, 2. PERT, 3. Precedence Diagramming,
4. Node Diagramming, 5. GERT, and 6. Bar Charts. Index these alternatives
by (i), so i = 1, 2, . . . , 6, in this illustration.
Some of the criteria to consider in choosing a networking system include:
1. Ease of drawing the initial network. 2. Ease of obtaining the initial data,
. . . , 9. Ease of updating effectively. Index these criteria by j = 1, 2 , . . . , 9.
Now assign weights, wi,t o each criteria t o reflect their relative importance.
Finally, for each network system (i), assign a relative score, sii,t o reflect how
it achieves the jth criteria. The final score for each networking system is then
computed from
9
Score for System ( i ) = Si = wjslj
i=l

Solution unacceptable, redefine problem

Project [ Problem Research Evaluate \ 1 i r i t i cype implementation

FSource

Redefine problem
Project washout /
382 11 / Advanced Topics

Chapter 11
1. Market Survey Project Effective Date: 8

Est. Act.
I J Our. Our. Description ES LS EF LF F

Plan Started Finished 3

Design Started 11 9 -2
Hire Started 10 10 0
Train 11 18 17 -1
Dummy 11 11 10 -1
Select 11 15 17 2
Print 11 19 17 -2
Dummy 19 19 17 -2
Conduct 19 34 32 -2
Analyze 34 39 37 -2

a. No. It is two days behind schedule.

b. 2-4-6-5-7-8
c. I . Updating feature for actual start and finish dates.
2. Activity revision feature.
3. Use of actual start and finish times for schedule computations

The computed schedule is the same. The input format would have one iden-
tifying number for each activity, rather than a pair of numbers (I and J). The
input would also list the dependent activities for each activity.
5. Use of a computer would be recommended because of the size of the net-
work, the number of updating calculations projected, and the various output
sorts desired. Software features needed would include: updating, cost sum-
 Solutions to E*orciws 383

marization, graphic output, calendar dating, and sorting by responsibility

code. Yes, several of the listed programs would be insufficient.
7. The contractor could use manual processing during the detailed planning
phase of the project, in order t o develop the most efficient sequence of activ-
ities for constructing each floor. He could then resort to a simpler network
for monitoring the actual construction progress.
INDEX

Abdelnour, J., 300, 311 Battersby, A,, 80, 114

Activity B e ~ i n g s o nL.,
, 184,186, 187
accounting, 136 (see also cost mntrol) Bensen, L. A., 252
definition, 23 Berkwitt, G . I., 20
dummy, 24 Beta distribution, 283
times (see time estimation) Beutel, M. L., 163
Activity duration estimation (see Time Bird, M. M., 325
estimation) Booz, Allen and Hamilton, 12
Activity time-cast curves, 240-243, 265, Budgets, 135
268 budgeted cost schedule curve, 144,145
feasible time-cost points, 267 Burgess, A. R., 204,225
pseudo activities, 242
Adamiecki, K., 9.10 Central limit theorem, 276-280,286
Agin, N., 225 Chaker, A. A., 300,311
Ang, A. HS., 300,311 Charnes, A., 252,257
Archibald, R. D., 163 Clark,C. E., 20,283,311
Arisawa, S., 322.333 Clark, R. A., 334
Clayton, E. R., 252,253,264,325,334
Backward pass (see Computations on arrow, Cleland, D. I., 8,19,185-187
node or precedence diagrams) Computations on arrow diagrams
Bar chart '
float (slack)
comparison to networks, 5 , 6 activity (free), 79, 85
computer produced, 107,113,350,354, allocation of (see resource scheduling)
356 independent, 81-82
preparation of, 5 interfering, 80-81
uses in resource allocation, 195 total (path), 78,85
386 Index

Computations on arrow diagrams (cont.) applications, 18

forward and backward pass, 72.76 history, 13,14
using linear propramming, 257-261 procedure, 239,241,243
using special symbols, 82-85 Craig, J. D., 38,94
using tabular ~rocedures.121-123 Crandall, K.C., 68,94, 105,114,121,272,
with multiple start and end events, 307,311
87-89 Criticality of activities
with scheduled dates, 87,90 in PERT simulation, 306,308,309
tabular procedures, 121-123 in precedence diagramming, 99
updating, 91-93 Critical path (see also Computations on
Computations on node diagrams, 93,94 arrow, node or precedence diagrams)
for resource allocation, 208-211 augmentation of activities on, 243,245
Computations on precedence diagrams criticality index from simulation, 306
anomolies, 96 definition, 80,86
assumptions, 101 from forward pass only, 87
computer outputs, 106, 107, 112, 113 in precedence diagrams, 99,100
constraint definitions, 94,95 reduction of duration, 238
critical path types and classification, 99, Cukierman, A., 248,250,252
100
no splitting allowed algorithm, 102 Davis,E. W., 20,68,202,225,228,238,
splitting allowed algorithm, 128-130 25 2
splitting of activities, 98,126 Davis, G , 187
continuity of work flow, 108, 127 DDR International, 175
Computer applications Department of Defense. 138
advantages, 337 WD/NASA Guide, 138,140
factors to consider, 338-340 DOD C/SCSC implementation guide, 140
input, 340-344 Digman, L. A., 333,334
output, 295,331,332,343-347 du Pont de Nemours & Co., 13
software, 340,348-370
software vendors, 148,368-370 Eisner, If., 321,333
Computer programming basic scheduling Elmaghraby, S. E., 20,225,252,263,321,
computations, 120-125 322,333,334
Contractual requirements, 166-168 Event definition, 24
Cooper, W. W., 252,257
Cost control, 133-161 Fazar, W., 20,269,311
ARTEMIS system, 148-161 Fendley, L. G., 225
basic problems, 136 Float
by computer, 151-161 activity (free), 79,85
computer software, 356,364-367 effect on limited resources, 192
cost plan or budget, 134,135 independent, 8 1,82
C/SCSC, 140 interfering, 80,81
PERTICost, 138 total (path), 78,85
Costs zero slack (float) convention, 76
cost measurement concepts, 143-146 Fondahl, J., 14,20,37,38,43,94,114,
cost performance index, 15 1 253
cost variance, definition, 144 Ford, L. R., 257,263
crash time-cost, 241 Forward pass (see Computations or arrow,
direct activity costs, 237, 240 node or precedence diagrams)
indirect project costs, 237, 240 Fox, 1.,139,140,163
CPM (Critical Path Method-see also Time Freeman, R. I., 321,334
cost trade-offs) Fulkerson,D. R., 114,122,253,257,263
 Index 387

Galbraith, I., 180, 183, 187 Marshall, A. W., 19

Gantt chart (see Bar chart) Martin,C.C., 180, 182, 183,187
GERT (Graphical Evaluation and Review McDomeU Automation Co., 163
Technique) Merge event bias (see PERT probability)
illustrative applications, 325-333 Metier Carp., I48
nomenclature, 322, 323 Meyer, W. L., 253,268
Gomez, R. S., 334 Moder, J. J., 20,114,282,283,312,333,
Gordon, J. H., 225 334
Moeller, C. L., 333, 334
Hald, A , 315,316 Montalbano, M., 115,120
Halpin, D. W., 333,334 Monte carlo simulation,
Hanscomb Assoc. Inc., 175-178 activity criticality, 306-308
Happ, W. W., 334 confidence intervals on results, 307-310
Harmanygraph, 10-12 on CERT, 330-332
Herbert, J. E. 111,322, 334 on PERT, 305-307
Heidorn, C. E., 225 sample size determination, 307-310
Hindelang, T. J., 253,268 Moore, L. J., 252,253,264,325,334
Huber,W. D.,217,218,226 Mueller-Mehrbach, H., 225
Multiproject scheduling, 194, 195
Implementation, 168-170 Muth, M. R., 253, 268
construction industry, 174-178
NASA Guide, 138
Johns Hopkins U., Applied Physics Lab, 294 Navy, 12
Johnson, T. J. R., 225,232 Networks
Jones, P. C., 187 arrow scheme, 25,63-67
computer produced, 360-362
Kelley, J., 14,20,237,238,253, 362 condensation, 58-60,355
Kendall, M. G., 311 constraint types, 40,95
Kerzner, H., 180, 182, 183, 187 definition, 25
Khumawala, B., 228 event scheme, 293,294
Killebrew, J. B., 225 generalized schemes, 321
King, W. R., 8,9,311 GERT scheme, 321
history, 5,10
Lawrence, P. R., 186,187 level of detail, 60-63,168
Lee, S. M., 252, 253,264 natural dependency, 34
Limited resource scheduling application, node scheme, 22,37,38,93
217,218 node types, 323
Linear programming, precedence diagramming, 22,38,63-67,
basic scheduling computations. 257-261 94
network flow theory, 263 resource dependency, 34
time-cost trade-off applications, 261-268 rules, 25,26,94,95
Lockheed Aircraft Co., 12 time-scaled, 36,37, 172-174
Lockyer, K. G., 12,19 Nomenclature, vii, viii
Long, J. M., 187 Normal distribution,
Looping, 320,321 discussion of, 277
Lorsch, J. W., 186, 187 tables of, 315,316
uses of in PERT, 286,301
MacCrimmon, K. R., 291,296,311 Normal time-cost, 237,240
Mahler, P., 370
Malcolm, D. G., 20,269,311 Optimization procedures (see also Linear
Mark, E. J., 68 programming)
Optimization procedures (mnt.) Remington Rand Univac, 13
branch and bound, 218,228-236 Resource allocation,
enumeration techniques, 218 computer graphics, 357,358
for limited resource scheduling, 217-219 computer software, 363,364-367
Organizational breakdown structure, 142 Resource scheduling, 202 (see also Optimi-
use in cost control system, 150,153, 156 zation procedures)
basic procedure overview, 202
Panagiotakopoulos, D., 253,268 heuristics, use of, 202,208,210,212,
Patterson, J. H., 217, 222,225, 226 216
Peck, M. J , 20 leveling resources, 204-207
PERTICost (see Cost control) parallel vs. serial methods, 203
PERT probability (see also Probability, and use of network characteristics, 220
Time estimation) Rodgers, E. G., 282,283,312
conventional, 271 Roseboom, J. H., 20, 269, 311
in computer software, 357, 359, 362, Rubin, I. M., 6 8
364-367 Ryavec, C. A , 291,296,311
merge event bias, 296,300
overview, 270 Sayer, J. S., 20
Peterson, P., 370 Scherer, F. M., 20
Phillips, C. R., 370 Scbrage, L., 226,232
PNET Seelig, W. D., 6 8
description, 302 Seiler, J. H., 187
illustrative application, 303-309 Sewall, R. F., 252
modified procedure, 303-305 Shaffer, L. R., 253,268
Polaris program, 12, 18 Shiffer, Z. F., 248, 250, 252
Precedence diagramming, 94, 175 Siemens, N., 245,246,253,263
computer software, 364-367 Simulation (see Monte carlo simulation)
Pritsker,A. A. B., 321, 322, 334 Slack (see Float)
Probability Smith, L. A., 370
basic, 272 Splitting of activities, 98, 126
conditional, 288 Standard deviation
cumulative, 288,301,309 definition of, 276
distributions, 273-275 of activity duration, 283
of meeting scheduled date, 286 of event times, 284
Probability branching, 320,321 of project duration, 286,290,295,301
Project Stinson, I.P., 219,220,226,228,232
control, 17, 170-174 Stochastic networks (see GERT)
definition, 3 , 4 Stuart, A. S., 311
planning phase, 21, 22 Stuckenbruck, L. C., 183, 187
proposals, 165,166
Project management Tabular procedures
advantages, 1 9 forward and backward pass, 121-123
definition, 3-4 for PERT probability, 304
human and organizational issues, 178- for time-cost trade-off, 245
187 Talbot, F. B., 217,226
methodology, 14-18 Taylor 111, E. R., 252,253,264
systems approach, 8-10 Time-cost curves (see Activity time-cost
uses, 18, 19 curves)
Project Managcment Institute, 368 Time-cost trade-offs
Prostick, J. M., 370 approximation hand procedure, 245-248
Pulat, P. S., 252 basic problem, 237-239
 computer software, 364-367 U.S. Army Corps of Engineers, 17, 187
definition of time-cost points, 240-241
economic perspective, 248
linear programming formulations, 261, Value of work performed, 143
263-268 Van Slyke, R. M., 305,307,312
minimum total cost schedule, 244,245, Vanston, J. H. Jr., 333,334
249 Variance (see also Standard deviation)
the CPM procedure, 239,241,243 cost variance, 144
Time estimation schedule variance, 144
accounting for weather, 52
accounting for weekends, 53
accuracy, 47-50 Walker, M. R., 20,237,238,253, 362
fixed time activities, 51 Weist, J. D., 99,115,214,226
monitoring, 292 West, 0.E., 370
Whitehouse, G. E., 334
most likely time, 281,284
Willie, C. T., 207, 226
optimistic time, 281
Wilson,T. A,, 311
pessimistic time, 281
sources of error in, 291 Wolfe, P. M., 333, 334
three estimate system, 280 Woodworth, 8. M., 207,226
Woopert, B., 371
use of historical data, 280, 317
Work breakdown structure, 138, 140,142
Time only computations (see Computations
Work package, 141
on arrow, node or precedence
diagrams)
Topological order, 101, 121 Yen, J. Y., 246
Turban, E., 20 Youker, R., 183,187