Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Philips

Download as pdf or txt
Download as pdf or txt
You are on page 1of 55

Eindhoven University of Technology

MASTER

Designing a metrics model for DevOps at Philips IT

Kruis, S.

Award date:
2014

Disclaimer
This document contains a student thesis (bachelor's or master's), as authored by a student at Eindhoven University of Technology. Student
theses are made available in the TU/e repository upon obtaining the required degree. The grade received is not published on the document
as presented in the repository. The required complexity or quality of research of student theses may vary by program, and the required
minimum study period may vary in duration.

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain

Take down policy


If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately
and investigate your claim.

Download date: 21. Nov. 2017


Designing a metrics model for DevOps
at Philips IT
Sander Kruis

Master Thesis

Eindhoven University of Technology


Eindhoven, November 19, 2014

Supervisors:
dr.ir. J.J.M. (Jos) Trienekens
dr. D. (Dirk) Fahland
Bas Vermeer (Philips)

1
Contents
1. Introduction .......................................................................................................................................... 4
2. Research model .................................................................................................................................... 5
3. Theoretical background ........................................................................................................................ 6
Agile Software Development .................................................................................................................... 6
DevOps ...................................................................................................................................................... 7
Lean Software Development .................................................................................................................... 8
Metrics ...................................................................................................................................................... 9
4. Methodology ....................................................................................................................................... 14
5. Current situation (As-Is) ...................................................................................................................... 17
IT Delivery ............................................................................................................................................... 17
Infrastructure & Operations ................................................................................................................... 18
Linking IT Delivery and I&O ..................................................................................................................... 19
Processes................................................................................................................................................. 19
Other observations ................................................................................................................................. 21
6. Design.................................................................................................................................................. 23
Goal definition ........................................................................................................................................ 23
Creating a reference framework ............................................................................................................. 24
First design .............................................................................................................................................. 24
First feedback round and second design ................................................................................................ 27
Second feedback round and third design ............................................................................................... 30
7. Discussion............................................................................................................................................ 32
8. Conclusion ........................................................................................................................................... 33
References .................................................................................................................................................. 34
Appendix A .................................................................................................................................................. 37
Original project description .................................................................................................................... 37
Appendix B .................................................................................................................................................. 39
Philips Integrated Landscape: Information Management ...................................................................... 39
Appendix C .................................................................................................................................................. 44
VIPP ......................................................................................................................................................... 44
Appendix D .................................................................................................................................................. 49
Lean measurements (Duque & Cadavid, 2007) ...................................................................................... 49

2
Appendix E .................................................................................................................................................. 50
Feedback session 1 (Round 1) ................................................................................................................. 50
Feedback session 2 (Round 1) ................................................................................................................. 51
Feedback session 3 (Round 2) ................................................................................................................. 52
Appendix F .................................................................................................................................................. 53
1st Feedback round................................................................................................................................. 53
2nd Feedback round ................................................................................................................................. 54

3
1. Introduction

Philips IT (Appendix A) is a centralized IT organization servicing three sectors of Philips; Healthcare,


Lighting and Consumer Lifestyle. Within IT there exist two large parties. The first is IT Delivery where
projects are executed and developed. The second, IT Infrastructure & Operations (I&O) is responsible for
daily operations. Daily operations include keeping the IT systems running and providing direct support.
IT Delivery has been adopting SCRUM to become more agile with over 100 SCRUM teams. These teams
are multidisciplinary and include relevant partners. These partners are located across the world, thus
these teams work together in a virtual way.

While IT Delivery is adopting agile methodologies, I&O is working according to an Information


Technology Infrastructure Library (ITIL) framework. While the organization is constantly changing, I&O is
looking for ways to become more agile with several teams adopting methods like SCRUM or Kanban.

Looking at the two parties, we initially see that IT Delivery is pressing for faster software releases as
SCRUM cycles are currently two weeks long. From operations however, system stability is of the highest
importance and releases are planned monthly.

IT delivery and I&O are now working together to improve this situation and release more often without
compromising on the quality of these releases. Since 2009 (DevOps, 2014), a method called DevOps was
introduced that focuses on the way Development and Operations work together. This method originates
from lean methodologies and stresses the improvement of flow, communication, feedback and
automation. Flow describes how work moves through the process. For example, increasing the speed or
decreasing the waiting times of work increases the flow.

Philips IT is looking to implement DevOps strategies, starting with two teams within IT delivery. On their
way to implement DevOps it becomes important to validate they are actually doing DevOps and how
this implementation process is affecting the organization. For this purpose a set of metrics will be
developed to measure whether DevOps is being achieved and how this can be seen within the
organization.

To develop these metrics there are a few questions that need to be answered with the main research
question being; How can DevOps be measured? To come to a solution for this question, we need several
other parts. There are two questions that immediately come up. The first being, what is DevOps
supposed to be? And secondly what has Philips IT done so far to implement a DevOps solution?

If we have an understanding of DevOps, and we know how Philips IT is currently working, we can work
towards a design to measure DevOps within Philips IT. After that we can follow a PDCA cycle to adapt
and improve that design.

4
2. Research model

Looking at the research questions we can see the path depicted in Figure 1.

Conceptual model

Theory
Software Design metrics
Validation
Development & model
Operations

Philips
Current situation

Figure 1 Research model

This path starts with background research on software development techniques and metrics related to
thes areas. From this theory some insights are gained from the different methods available. These
methods will include DevOps and SCRUM, as well as look at lean development. This is followed by
research on available metrics and how metrics can be designed. From this research a conceptual model
can be created, which depicts what a DevOps metrics model would entail. Looking at this conceptual
model as well as the current situation within Philips IT an initial model is designed, which is validated
and adapted several times in a single case-study. Designing the metrics model is done together with
people involved in the process in a form of action research.

5
3. Theoretical background

In this chapter several software development techniques are described to provide insight in the
terminology and provide a path towards a conceptual model. The first subject to look into is agile
software development, that being the type of methods that IT Delivery is currently working with as well
as what IT Operations is currently looking to implement. After that, a closer look at DevOps is provided,
corresponding with the goal of the research. From DevOps, some more research follows corresponding
to what DevOps is about, leading into a description of lean software development. This is followed by a
description of available metrics that are being used in agile and lean areas to provide a reference
framework while designing the final metrics model. Finally, some guidelines are described for ways to
design metrics.

Agile Software Development


The term Agile Software Development originated from the ‘The Agile Manifesto’ (Beck, et al., 2001) and
consists of several values and principles for faster and better software development. The four values
they describe are:

 Individuals and interactions over processes and tools


 Working software over comprehensive documentation
 Customer collaboration over contract negotiation
 Responding to change over following a plan

While there is not a single definition of agility, most incorporate the idea of adaptability to the
environment and quick value creation. One of such definitions is given by Erickson et al. (2005); “agility
means to strip away as much of the heaviness, commonly associated with the traditional software-
development methodologies, as possible to promote quick response to changing environments, changes
in user requirements, accelerated project deadlines and the like.”

While this definition is focused on software development, similar trends have been previously seen in
other disciplines. Conboy and Fitzgerald (2004) for example related agility to “flexibility” and “leanness”
as seen in agile manufacturing and lean development. However, several differences exist between the
terms. According to Zhang & Sharifi (2000) agility consists of two components, these being flexibility and
speed. Hereby stating that flexibility alone is not enough to be agile. Comparing agility to leanness,
complements each other with regard to simplicity and quality, but approaches economy differently
(Young, Muchlhaeusser, Pigging, & Rachitrangsan, 2001). While leanness attempts to remove ‘waste’
entirely, agility removes waste only to the extent that it does not hinder the ability to change, or be
flexible.

While the above definitions describe the values for agile software development, it does not provide a
direct method to apply these values. While a multitude of methods have been developed, a few of these
have become more prominent and are further described below in Table 1.

6
Table 1 Agile methods
Agile method Description
Dynamic Systems Projects are divided into three phases; pre-project, project life-cycle, and
Development Method post project and has nine principles for development being: user
(DSDM) (Stapleton, 2003) involvement, empowering the project team, frequent delivery, addressing
current business needs, iterative and incremental development, allow for
reversing changes, high-level scope being fixed before project starts, testing
throughout the lifecycle, and efficient and effective communication.
Scrum (Schwaber & The development is organized in sprints (short iterations of about 2 to 3
Beedle, 2001) weeks) by a self-organizing team. Each sprint goes through planning, design,
testing and review. Features that need to be developed are stored in a
‘Backlog’ where the product owner decides the items that will be worked
on in the following sprint.
Extreme Programming Focuses on best practice and consists of twelve practices: the planning
(XP) (Beck & Andres, game, small releases, metaphor, simple design, testing, refactoring, pair
Extreme programming programming, collective ownership, continuous integration, 40h week, on-
explained: Embrace site customers, and coding standards.
change, 2000)
Lean software Adapted from lean development and based on seven principles: remove
development waste, amplify learning, decide as late as possible, deliver as fast as
(Poppendieck & possible, empowered teams, build integrity, and see the whole picture.
Poppendieck, 2003)
Kanban (Anderson, 2010) Kanban is based on the theory of constraints (Goldratt & Cox, 1985) and
comes with six core practices; visualize, limit work in progress (WIP),
manage flow, make policies explicit, implement feedback loops, improve
collaboratively & evolve experimentally.

DevOps
While development and operations departments have different goals or key performance indicators,
DevOps attempt to satisfy both goals. As development wants to deploy more and more often,
Operations might want the exact opposite as in order to keep all systems running and stable it might be
more beneficial to not deploy anything new at all. As organizations have seen the need for agility, the
option to not deploy anything is not realistic at all, as customers most likely want new products or
features on a regular basis.

To achieve this agility, not just development should be agile, but the entire organization. DevOps
attempt to do this with three core ideas (Kim, Behr, & Spafford, The phoenix project: A novel about IT,
DevOps, and helping your business win, 2013) (Kim, 2013):

Table 2 The Three Ways of DevOps


1. Systems thinking Stresses that it is more beneficial to look at the performance of the
entire system, than the performance of specific parts of that system.
2. Amplify feedback loops Allows understanding of the customer and availability of knowledge
where it is needed.
3. Culture of continual Experimentation and learning helps you in the long-term by being able
experimentation and to more quickly adapt and respond to changes or problems.
learning

7
Systems thinking refers to an approach that looks at problems in relation to the entire system it exists in
(Ackoff, 1971). The approach ensures that the performance of the system as a whole is more important
than the performance of separate parts of the system (IT Delivery or I&O). Like in lean development the
performance described here is the performance related to the value of the system. Implementing the
first way leads to a few outcomes:

 Never passing a defect to following work centers


 Do not allow local optimization if it creates global degradation
 Always look to increase flow

Amplifying feedback loops leads to early knowledge of issues and problems, so the system can quickly
adjust where needed. Implementing the second way leads to the following outcomes:

 Understanding and responding to all customers


 Shortening and amplifying feedback loops
 Knowledge available where needed

The third way mostly supports the other two ‘ways’ to ensure that improvement is a continuous process
and it leads to:

 Making time for the improvement of daily work


 Reward teams for taking risks
 Introducing faults to increase resilience

Lean Software Development


Leading from systems thinking, as a method that looks at the entire value stream, Lean software
development is the application of lean principles and methods to the software development domain.
Lean focuses on the creation of value and the elimination of waste in the production process and was
first introduced in “The machine that changed the world: the story of lean production” (Womack, Jones,
& Roos, 1990). Some changes were needed to translate from production to software development
however. As Jailia et al. (2011) described lean software development is based on five principles:

 Specify value
 Identify the value stream
 Create flow
 Customer pull
 Pursue perfection through continuous improvement

In order to identify the value stream and create flow, lean focuses on eliminating waste. For software
development projects seven types of waste have been identified (Ikonen, Kettunen, Oza, &
Abrahamsson, 2010) (Poppendieck & Poppendieck, 2003):

 Partially done work


 Extra processes

8
 Extra features
 Task switching
 Waiting
 Motion
 Defects

Metrics
The next paragraphs describe metrics that have been used in areas related to DevOps. Starting with
measurements in lean manufacturing and scrum, which is mostly related to systems thinking and slightly
to feedback loops. Followed by a description of culture of learning and finally about how metrics can be
designed.

The first of these areas is lean manufacturing. Duque & Cadavid (2007) gathered a collection of metrics
that could reflect the advancement in the implementation of lean manufacturing. They used a collection
of case studies and gathered all metrics used in these cases. While not every metric is as useful as the
next, it does provide a good overview of what metrics are being used in practise. These metrics are
divided into five dimensions, in a similar fashion to earlier research by Karlsson & Ahlstrom (1996) and
Martinez & Perez (2001), but adapted for the goals of Toyota Motor company, for whom Duque &
Cadavid (2007) performed their research. The five dimensions that were used are; Elimination of waste,
Continuous improvement, Continuous flow and pull driven systems, Multifunctional teams, and
Information systems. Below is adescription of the metrics they found for these dimensions. An
additional overview of these metrics can be found in Appendix D.

For the first dimension of ‘elimination of waste’, Duque & Cadavid (2007) describe metrics related to
what kind of activities time is being spent on within the process. This describes activities like Setup time,
Downtime and Transportation. Along with these time allocations, they mention Space utilization and
Work In Progress (WIP). The last two metrics named are Work in Progress (WIP) and space utilization.
Although WIP in IT is usually not something tangible as WIP in a development and deployment process
mainly consists of lines of code some form of WIP could still be used. For example, scrum (Schwaber &
Beedle, 2001) and many other agile methodologies refer to work as user stories, which are small
increments of work that in itself have some value in the final product. The amount of user stories and
their relative size could provide a good overview of the size of WIP.

For the second dimension, ‘continuous improvement’, several metrics are are again described. These
metrics are related to either suggestions made by employees and how many of these get implemented
or the amount of rework that has to be performed. For the amount of rework two metrics are
mentioned. The first is the actual percentage of units that requires rework. The second refers to ‘scrap’,
which they define as the percentage of products that need to be scrapped, thus cannot be reworked
properly and are discarded.

For the third dimension, ‘continuous flow and pull-driven systems’, there are again several metrics
described. The first metric is the lot size. Translating lot size to IT development could refer to the
number of user stories that go through the system simultaneously, so for example the average amount

9
of user stories per sprint or the amount that simultaneasly get deployed. The second and third metric
refer to the flow time and lead time of an order. The lead time refers to the average time from the
creation of an order to the delivery of the order. The flow time on the other hand refers to the time it is
being processed. This thus does not include waiting or setup times. The last metric for this dimension
deals with measuring the pull-driven system, which is measured by the percentage of processes that pull
inputs from their predecessors and the value of their throughput. This would mean that teams that are
located early in the process do not simply work up to their capacity, but only as much as needed to
properly supply the following teams.

The next dimension deals with ‘multifunctional teams’. This relates to the flexibility of the team to do
each other’s tasks and how much responsibility they take for the product they deliver. The first metrics
for this dimension are the percentage of quality inspection the team performs themselves and the
percentage of tasks that are performed within the team as opposed to other teams. These together
describe how much influence the team itself has on the product that is delivered and how much the
team depends on the work of other teams. Other metrics that are mentioned are ‘Cross training’ and
‘Number of employees capable of assignment rotation’ (Duque & Cadavid, 2007). These metrics
describe how well the team members can perform each other’s tasks as well as how dependent the
team is on each member being present.

For the last dimension, ‘information systems’, the metrics relate to the availability of information within
the team and the documentation of procedures. The mentioned metrics however do not deal with the
quality of the provided information. The impact of the quality could of course change the measurements
done in the other categories. For example, if a team member would not have access to necessary
information, this would directly influence the flow time and would be adressed in the daily meetings
that are already in place. The metrics named are “the frequency with which information is given to
employees”, “percentage of procedures documented” and “the frequency with which the progress
boards are updated.” These progress boards refer to kanban-like boards that display the work in
progress.

Other metrics related to software development can be found within scrum research. One such research
by Downey & Sutherland (2013) describes ten metrics to promote improvements. These metrics are
described below in Table 3. While these metrics seem very different to the metrics described by Duque
& Cadavid (2007), the reason for that is that the object they measure is different. While the metrics by
Duque & Cadavid (2007) look at the entire value chain according to lean principles, the metrics by
Downey & Sutherland (2013) are used to measure the performance of scrum teams on their own. As
scrum teams usually perform the development before work moves to operations, these metrics thus
measure only a part of the value chain.

10
Table 3 Ten essential metrics (Downey & Sutherland, 2013)
Metric Description
Velocity The sum of the estimates of all accepted work.
Work Capacity The sum of all work reported during a sprint.
Focus Factor Velocity divided by Work Capacity, described how much time was spent on
planned work.
Percentage of Describes how much extra work the team has done. (Adopted work divided by
adopted work original planned forecast)
Percentage of Describes how much extra work the original work has cost. (Found work divided
found work by original planned forecast)
Accuracy of Describes how well the forecast was estimated, thus how much time work took in
estimation contrast to the time planned to do it.
Accuracy of Describes in how far the forecasted work was equal to the actual work done.
forecast
Targeted Value Current sprint’s Velocity divided by original Velocity.
Increase
Success at scale Accepted attempts to scale divided by the total attempts to scale.
Win/Loss record Each sprint is a success if:
 A minimal of 80% of original forecast is accepted.
 Found plus adopted work remains at 20% or less of the original forecast

Metrics for the third way, a culture of continual experimentation and learning, are lightly found within
the second dimension of ‘continuous improvement’. However, some further research can be performed
here. Research by López, Peón and Ordás (2004) provides a ‘definition’ of organizational learning. While
not directly providing metrics, this definition helps the reference framework by better understanding
what organizational learning is composed of. Their research defines four different constructs of
organizational learning.

 Acquisition of knowledge, either through external sources or internal development


 Distribution of knowledge
 Interpretation of knowledge or how the organization shares and uses knowledge
 Organizational memory or how the organization stores the knowledge

To design metrics, Pfleeger (1993) describes some lessons learned in developing a metrics program at a
telecommunications firm called Contel. The main lesson there is that “Metrics are welcome only when
they are clearly needed and easy to collect and understand.” Besides this general statement Pfleeger
describes ten lessons, which can be seen below in Table 4.

11
Table 4 Pfleeger's lessons learned (Pfleeger, 1993)
Lesson Description
Begin with the process The process helps finding problems and identifying what the metrics should
be used for.
Keep the metrics close to Develop the metrics with the people who will work with them. This leads to
the developers better understanding.
Start with people who Start with some success stories, which will spread themselves. Thus makes
need help; then let them other teams follow more quickly.
do your advertising for
you
Automate as much as Minimize the time needed to collect data as it is more useful to spend time
possible analyzing the data than collecting it.
Keep things simple and Keep the metrics close to the problem and do not create too many metrics.
easy to understand
Capture whatever you Capture as much as possible without interfering with general work. Some
can without burdening things might be useful for other problems than the one you are currently
developers solving.
If the developers don’t The quality of the resulting data might not be very good otherwise.
want to, don’t make
them
Using some metrics is The general idea is to start with success stories. Measuring one or two
better than using no useful things might convince people other metrics could be worthwhile.
metrics
Use different strokes for Keep in mind the products, environments, domains, goals and customers.
different folks
Criticize the process and Focus the metrics on the process and/or product.
the product, not the
people

While these lessons provide some guidelines in how to proceed while designing metrics, they do not
provide a structured way to actually design them. For that the Goal-Question-Metric (GQM) approach
(Van Solingen & Berghout, 1999) can be used. This approach defines three levels for a measurement
model (Basili, Caldiera, & Rombach, Goal Question Metric Paradigm, 1994), namely; conceptual (goal),
operational (question), and quantitative (metric). Starting from a clear definition of the goal, followed by
questions that would describe the goal, metrics can be designed to measure the goals.

In closing, the background research has provided some insight in what DevOps is about, followed by an
overview of metrics that could provide useful for the design of a metrics model for Philips. Also some
guidelines for designing a metrics model have been shown, which will help describing the methodology
below. A summary can be found below in Table 5.

12
Table 5 Summary of background research
Subject
DevOps  Systems thinking
 Amplifying feedback loops
 Culture of learning and experimentation
Metrics  Lean manufacturing
 Scrum
Designing  Begin with the process
metrics  Keep the metrics close to the developers
 Keep things simple and easy to understand
 Goal-Question-Metric apprach

13
4. Methodology

This chapter describes the methodology used in this research.

As some background knowledge on related topics is now gathered, DevOps can be described in terms of
a conceptual model. This model depicts what factors DevOps is composed off. From research in Chapter
3 we see DevOps is composed of three ‘ways’, namely; systems thinking, feedback loops and culture of
learning & experimentation. These three can thus be seen in the conceptual model (Figure 2). When we
can measure these three ‘ways’ through the metrics model to be designed in this research we can see
the performance of DevOps withing Philips IT. After that, the model could lead to new insights to adapt
the model.

Philips IT

DevOps

Systems
thinking

Feedback Metrics
loops + model

Culture of
Learning &
Experimentation

Figure 2 DevOps conceptual model

To design this metrics model, some steps will have to be taken. This starts with getting a better
understanding of the environment the model will be used in. This means a description of the current
situation within Philips IT is needed. To do this properly, a general introduction of the environment is
provided first. After which follows a more in-depth look into how the three ‘ways’ can currently be seen
within Philips IT. For this, two teams within Philips are available to gain this understanding. To provide
more understanding of especially systems thinking and feedback loops as well as following Pfleeger
(1993), the process these teams follow is mapped and compared. For the third way, culture of learning
& experimentation, a more general impression suffices, by describing the way the teams perform daily
activities.

Following the analysis of the current situation, a metrics model can be designed. Because of technical
difficulties, it was not possible to get data from the teams about designed metrics or metrics that were
already in place. For this reason, the model is designed together with relevant parties in the form of

14
action research (McNiff, 2013), where a solution is designed by (or with) those working in the
environment itself. Also, this ensures the metrics model will be better understood and closer to those
working with it afterwards along the lessons learned by Pfleeger (1993).

The goal of this research is to investigate the situation within Philips IT. Therefore a set of metrics will be
designed that together determine how well the Philips IT is doing in terms of DevOps. In order to come
to such a design, several steps have to be taken along the lines of the Goal-Question-Metric approach
(Van Solingen & Berghout, 1999). First the goal has to be further specified. This would describe what we
need metrics for exactly and provides a better insight in which way the design is moving. From there a
reference framework should be created. This is primarily done to provide some insight in what is already
present in today’s literature to further understand the goal. Following the literature review, a first
design can be created. To ensure the design is useful, feedback is gathered within Philips after which this
feedback is processed and the metrics are updated. During this feedback process the design is
constantly being validated.

Determine the goal

Create a reference
framework

Design metrics Collecting feedback Processing feedback

Conclusions

Figure 3 Design methodology

Figure 3 shows the steps taken in this research. The following paragraphs will describe how the different
steps are performed in more detail.

While determining the goal and designing the metrics are two separate steps, the Goal-Question-Metric
(GQM) approach (Van Solingen & Berghout, 1999) can be used in both these steps. This approach is
meant to discover which metrics are suitable for your problem by looking at the goal that the metrics
are supposed to be used for. The first part of this approach solely looks at the definition of the goal,
which is what is needed here. To help defining these goals a template is used. One such template is
suggested by Basili, Caldiera, & Rombach (1994) and is shown below in Table 6.

15
Table 6 GQM goal definition template (Basili, Caldiera, & Rombach, Goal Question Metric Paradigm, 1994)
Analyze The object under measurement
For the purpose of Understanding, controlling or improving the object
With respect to The quality focus of the object that the measurement focuses on
From the viewpoint of The people that measure the object
In the context of The environment in which measurement takes place

To create a reference framework, literature research was done to provide insight in what metrics are
currently available and used for related or similar goals. This research is focused on metrics that are
used for DevOps or similar situations. As metrics for DevOps were not readily available, a look into lean
manufacturing and scrum metrics was taken. While GQM technically states that the metrics should be
designed with just the goal in mind, those using GQM generally possess a reference framework of the
sort, due to them working in the particular environment. This step is there to cope with the lack of such
a reference framework for me personally. Also, it could be interesting to see whether following GQM
leads to similar metrics as the literature research suggests, or whether these will be different.

After a reference framework has been created, GQM can be further used to design a set of metrics
according to the already defined goal(s). The next step in the GQM process is to devise questions for the
goal(s). These questions should refine the goal to make it easier to approach. Good questions will also
tell you whether the goal is reached. To answer these metrics one or more metrics are suggested which
together form a metrics design.

For the feedback loop, several relevant parties are asked to provide feedback proposed metrics. The
feedback collected resembles an embedded case study design (Yin, 1989), where several parties in the
context of the case study are studied. This feedback is provided in an iterative manner, where feedback
provided is processed into the design, after which new feedback can be given. This ensures the design
will be more valuable later on, and those working with it have a better understanding of the design and
how it was created. Depending on the feedback this feedback loop can be applied multiple times. While
this could technically always continue, a stop-criteria is needed. As feedback is processed in each loop,
the involved parties should come to a form of agreement in a few cycles. Therefore whenever the design
is not significantly changed from its previous iteration, the cycle will be stopped.

Due to the process of designing the metrics model, a separate validation is not performed. Because of
the repeated sessions to gather feedback, the model is constantly being validated.

16
5. Current situation (As-Is)

As described in the introduction, Philips IT is divided in two large parties, namely Delivery and I&O. In
order to work together it is important to realize the places in which these parties have common ground
as well as where the differences lie. In order to gain insight into this, both parties are described in the
next paragraphs. After a description of these two parties, the process in which they interact is described,
followed by some final observations.

IT Delivery
Philips IT Delivery is the party responsible for creating and acquiring IT solutions. The solutions are then
passed on to I&O for maintenance and support. As Philips has a wide array of IT services, IT Delivery is
divided into several delivery centers based on the end to end value chain (Philips IT, 2013), namely:

 Idea to Market (I2M)


 Market to Order – Digital (M2O – Digital)
 Market to Order – Customer Relationship Management – CRM (M2O – CRM)
 Order to Cash (O2C)
 Information Management (IM)
 Integration
 Enabling Processes (Human relations, Finance, Procurement)
 Common Infrastructure & Operations

Figure 4 describes shows these eight delivery centers to provide some background on where they exist
in the end to end (E2E) process.

Figure 4 Philips Integrated Landscape E2E process

IT Delivery has been adopting Scrum to become more agile, along the lines of the Scaled Agile
Framework (SAFe) (Leffinwell, LLC., 2014). This framework described three layers. The top layer is called
the portfolio layer, which describes the alignment to the business and mainly consists of epics. These
epics are large business needs or goals. For IT there are two different types of epics specified, namely;

17
Business epics and Architectural epics. The first, business epics are customer focused and deliver direct
value to the customer and thus the business. In contrast, the architectural epics are more supporting
needs that ensure the ability to develop and support future needs.

The middle layer is called the program layer, which describes the alignment of agile teams that each
deliver parts of the above epics. These epics are further specified here into features that can be
developed separately.

The bottom layers, the team layer, describes the workings of the actual development teams. The
features are describes further into user stories. These user stories are work items that can be completed
within a specified sprint (which is usually two weeks). The teams work on these user stories and
complete an amount of these within each sprint, so that at the end of each sprint some working
software/development can be delivered.

The scrum teams within Philips IT are Multidisciplinary teams (MDTs). This means that there is regular
contact with every stakeholder of the project they are working on. For example, such a team consists of
developers, testers, a product owner, a relevant partner. Also, there is an Infrastructure and Operations
Manager (IOM) within these teams, who exists as a link between these MDTs and IT Operations. These
teams hold short daily meetings to discuss the current status and issues of their current project and the
way they perform their work.

Infrastructure & Operations


While IT Delivery is responsible for developing new solutions, I&O is responsible for maintaining the
existing systems as well as providing direct support to customers.

While process-wise work flows from Delivery to Operations and it would be beneficial to work together,
historically these two parties are quite far apart. With work figuratively being thrown over a wall to
Operations by Delivery. While this does not directly seem to be the case within Philips, with the IOMs
within the delivery teams, there is still a clear gap between how both parties work. On the one hand
there is IT Delivery working according to SAFe, thus attempting to work agile, while on the other hand
I&O works according to ITIL (ITIL 2011 - The big picture, 2011). This way of working is more structured
and based on following clear processes than working in an agile way.

From ITIL there are several categories of work that exist within IT Operations. After initial meetings with
people working within I&O these categories have also been clearly visible.

Table 7 ITIL Types of work


Category Description (Nissen, Christian Feldbech; CFN
People, 2010)
Incident Management The process to restore service to users in case of
an incident, error or outage.
Service Requests The process of responding to and managing user
requests, for example password resets.
Change Management The process of applying changes to software or
hardware.

18
Problem Management The process of analyzing the root cause of
incidents or errors and for attempting to
determine where future incidents might happen.
Release and deployment Management The process of planning releases to environments
and ensure that the live service remains stable
during those releases.

While I&O is still working in this way, there are initiatives within Operations to become more agile. This
initiative is called AgileOps within Philips and is setting up teams similar to the MDTs within IT Delivery.
By looking at the above types of work several are more suitable for an agile way of working. These types
of work are mainly the problem management and the release and deployment management. The other
types of work are tasks that are in general very small and tasks that are unplanned. The two types that
are suitable are types that can be planned and worked on at any time.

Linking IT Delivery and I&O


While both parties have different types of work and their way of working is quite different, these two
parties do have to work together in some areas. Probably all parts could benefit from working together
in some way to make sure IT Delivery accounts for issues Operations might encounter and Operations
delivers feedback when something does go wrong. However, there are parts that have a more direct link
than others. Traditional IT development depended on a method known as waterfall (Royce, 1970). This
method describes several steps for the development of software. In short, these consist of
requirements, analysis, design, development, testing and deployment. As scrum teams take care of all
these but the deployment into operations, this is where the most important link between IT Delivery
and I&O exists. Also, as DevOps focuses on quickly delivering and deploying IT solutions, this is the most
obvious focus to take.

Processes
To better understand the process of deployments, two teams were available. These teams, respectively
Philips Integrated Landscape: Information Management (PIL IM) and VIPP , are quite different. The VIPP
team is a settled team which focuses on development for legacy systems, while the PIL IM team is
relatively new and focuses on development for the new landscape (PIL).

Going from lean perspectives described in Chapter 3, the first step was to determine the value to be
delivered by these teams. Speaking to these teams, the value they deliver comes down to the service or
product that has to be deployed to the respective platform. While this is the initial value, the system will
also be supported and maintained, thus apart from the deployment and the value to the end-user, there
is also an internal customer, namely I&O. In the end, I&O is also delivering value in terms of support to
the end-user.

Both teams follow a process to deliver this value (Appendix B & Appendix C). This process however is
different for both teams. While the teams were working in these processes, there was no clear view of
this process and it therefore had to be created.

19
The processes were created over several review sessions where the process was discussed step by step
with those working within the teams. The feedback was then processed and reviewed again in the
following session. This continued until a point where the process was agreed upon and could provide
the members with a clear understanding of how the process works and who are involved. This does
mean the process is a perfect representation, as the goal here was to create an understanding of the
process and not to for example implement this process into a workflow system.

The process for the PIL-IM team is shown in Figure 5 and more closely in Appendix B. The process starts
where a feature is placed on the team’s backlog for a new Information Factory (IF). Before development
can begin, some preparation has to be done, where models and specifications are created. When this
has been created and approved the actual IF is requested. This practically means that if the request is
then approved the development can start. From there a partner of Philips, in this case Cognizant, starts
development and deploys the IF on the Dev environment. Here the IF will undergo unit testing before it
can be transported to the QA environment. In this environment a system integration test (SIT) will be
performed as well as a user acceptance test (UAT). From there some preparation has to be done to
prepare for the release of the IF as well as training and knowledge transfer to support teams. Finally the
IF can be deployed to the production environment, after which some final adjustment have to be made,
as well as loading the appropriate data. For some time the Delivery team stays responsible for early life
support before the final handover to I&O.

Figure 5 PIL-IM E2E process for the creation of an Information Factory (IF)

The process (Figure 6 & Appendix C) of the VIPP-HC team works similarly. It starts with a feature on the
team’s backlog. For this feature several user stories are created and some preparation is done before
development. When a user story is then picked up in a sprint, it gets developed and unit testing is done.
From there a SIT is performed on the Dev environment. If these tests are approved the development is
transported to QA and another SIT is performed followed by the UAT. If these are approved a final
regression test is done and preparation for deployment to the production environment starts. When this
has been done the deployment has to be approved and is then performed. Some checks in production
are done for the integrity of the system and data is loaded. Then when the release is accepted there is
the final handover to I&O.

Figure 6 VIPP-HC E2E process

20
While the processes of both of these teams are different there are some clear similarities between them
as well. Using these two processes a general flow can be deduced that both processes follow and shows
us in general steps what happens within these processes. Figure 7 shows this more general flow which
starts with a feature on a backlog that needs to be developed. When this item is picked up there is some
preparation that needs to be done, referring to the different specifications that need to be determined.
From there it is passed to the delivery team which does some additional preparation before actual
development begins. In the VIPP team, this first preparation is also done by the delivery team however.
After this has been done, the actual development begins. Development can take place within the team
or with an external party. After development, some testing is done, which includes at least a SIT and a
UAT. When these tests have been accepted the service or product can be deployed in the next release,
after which it can be supported by I&O.

Development by
Delivery Team

Feature on backlog Preparation Delivery Team Testing Deployment Live

Development by
external party

Figure 7 Delivery flow

To further describe the status of the processes, some data should be collected, relating to lean
measurements. That way bottlenecks and issues can be identified to provide a direction for a possible
redesign. Within Philips most user stories are kept within a system called Rally (Rally Software
Development Corp., 2014). However, due to technical difficulties, it was not possible to get this data.

The creation of these processes however, was a clear step towards systems thinking, as by
understanding the entire process it becomes more likely actions can be taken to improve the entire
process, opposed to just parts of the process. For feedback loops the amount of approval steps can be
gathered from the process itself. However, without data about process times, this in itself does not
provide much information.

Other observations
Besides the way both parties work as well as the processes they work with, there are a few other
observations that can be made. While these are not directly linked to DevOps, an impression about the
culture can be gained from these observations.

The first of these can be seen in the way sprints are handled. What scrum tells us is that work is divided
up into small user stories that each have value to the user or customer. Each of these user stories can
then be picked up within a two week sprint. Within this sprint the user story goes through the full design
cycle, meaning it will be dealt with in a small waterfall-like manner. Thus, a design is made, it is
developed, tested and preparations for deployment are made. However, what can be seen in the
processes is quite a linear flow. As well as from some indications by team members is that what is done
within a sprint is mainly the development and perhaps some unit testing. Besides this, it can also be
seen in the planning of sprints. This planning usually starts with a sprint that is dedicated to the actual

21
planning of the project. The next sprint will then deal with designing the solution, after which one of
more sprints are planned for the development. In case there are more sprints for this we can see
something that looks like actual scrum. Then after development there is a sprint for testing and then
finally there is a deployment. The only part of this that is actually scrum is perhaps the actual
development part. For the rest it looks exactly the same as a traditional waterfall method, with the
exception that every two weeks there is a report about what has happened in the previous two weeks.

The second observation relates to the PIL-IM team. At this time, the actual generation of the
Information Factory (IF) is done by an outside party, Cognizant. The agreement with the partner is that
whenever a request is made by Philips, cognizant has two weeks (sprint length) to deliver on this
request. This two week period holds for both small changes as well as an entirely new IF. What happens
here is that leadership encourages that the specification that is sent to Cognizant is 100 percent correct
(one-time-right). However, DevOps encourages a culture of experimentation and learning. This is
however conflicting with the one-time-right attitude. This approach makes sense when every change
takes two weeks, after which it is tested within Philips, and has to go through another two week cycle if
something is wrong. However, should this agreement be based on the actual amount of work needed
for the change, thus a more agile agreement, the focus does not have to be on one-time-right. This
could then considerably speed up the preparation phase as well as the development phase.

From the situation and observations described in the previous paragraphs, there are definitely things
that show DevOps thinking, however there are also observations that in a way contradict the message of
DevOps. There are three areas in which there could be some clear improvements. These areas are:

 Scrum and waterfall, whether real scrum actually exists or would be useful
o This mainly relates to the fact that the content of sprints is still planned like it is
waterfall
 Approach on agreements with partners
o While Philips is on its way to work in an agile manner, the most benefit can be gained
there if the partners are supportive on that. The agreements with partners should take
this agile way of working into account.
 Metrics on DevOps
o While systems thinking seems to already be more or less present in the teams I saw, it
would be beneficial for management and the teams itself to be able to see how they are
doing.

22
6. Design

As described in the methodology there are several steps to take to come to a final design. This chapter
starts with the definition of the goal(s). Following the goal(s) is an overview of research on possible
metrics to create a reference. From there the Goal-Question-Metric (GQM) method is followed to
determine which metrics are needed and most valuable for the purpose of DevOps. Finally some
feedback rounds are described where the metrics model is updated and validated.

Goal definition
The first step of the GQM process as well as the creation of the design is to define the measurement
goals. To help defining these goals a template (Table 8) by Basili, Caldiera, & Rombach (1994) is used.

Table 8 GQM goal definition template (Basili, Caldiera, & Rombach, Goal Question Metric Paradigm, 1994)
Analyze The object under measurement
For the purpose of Understanding, controlling or improving the object
With respect to The quality focus of the object that the measurement focuses on
From the viewpoint of The people that measure the object
In the context of The environment in which measurement takes place

The object in this case in the development and deployment process within Philips IT. This is the process
that has been described within Chapter 5. The initial purpose of this research was to design a DevOps
solution. While Philips has already started a pilot in this, the purpose for the measurement is to further
understand this process and if possible improve it. The focus would be on the three factors that DevOps
describes, thus systems thinking, feedback loops and a culture of learning and experimentation. To
discover the viewpoint, several stakeholders are involved in the process. There are three stakeholders
that can quickly be identified, namely, the Philips IT management, the customer of the process, and the
teams within the process (Delivery, test factory, I&O). As the purpose of this research as well as that of
the DevOps pilot within Philips is to discover whether DevOps is useful, the measurements are firstly
meant for the management.

This leads to the following goal: Analyze the development and deployment process within Philips IT to
further understand and improve with respect to the three ways of DevOps; systems thinking, feedback
loops and culture of learning and experimentation from the viewpoint of the IT management.

As the quality focus in this goal actually contains three different foci, these three could be seen as sub
goals. Thus, the first goal is to understand and improve with respect to systems thinking, the second
goal with respect to feedback loops and the third with respect to culture of learning and
experimentation. These goals can be seen below in Table 9.

23
Table 9 Goal descriptions
Goal 1 Analyze the development and deployment process within Philips IT to further understand and
improve with respect to systems thinking from the viewpoint of the IT management.
Goal 2 Analyze the development and deployment process within Philips IT to further understand and
improve with respect to feedback loops from the viewpoint of the IT management.
Goal 3 Analyze the development and deployment process within Philips IT to further understand and
improve with respect to culture of learning and experimentation from the viewpoint of the IT
management.

Creating a reference framework


Following the three goals, it becomes clear the reference framework should look into several different
metrics, namely; systems thinking, feedback loops and culture of learning and experimentation. Besides
this it is useful to look into metrics used in similar fields. This would be looking into metrics related to
lean manufacturing or Scrum. The description of these metrics can be found in chapter 3.

Along with the background on metrics mentioned in chapter 3, several additional factors must be
regarded. These factors mainly relate to the quality of work. In regard to development being mostly
knowledge work, doing work faster usually means a trade-off in quality. This trade-off in quality is
however not accepted by Philips. The metrics (Duque & Cadavid, 2007) do not directly measure quality,
however, the quality can be seen in the percentage of rework that has to be done and the amount of
quality control done by the team itself. Also, Philips I&O is already measuring the amount of incidents
that were raised following a deployment. This number can also be used as an indicator for quality.
Finally, it must be taken into account that there is not just a single team working on a platform. Again for
the identification of bottlenecks and issues the capacity of these teams could be looked at. Again, while
this is not measured directly, the Work in Progress (WIP) does show a part of this as more teams in
general allows for more WIP.

Using these metrics covers the first way of DevOps, along with a part of the third way. The second way is
also covered in the five dimensions as rework. This would however only include negative feedback and
not necessarily the positive feedback. While feedback is generally also provided in the daily team
meetings, related to agile development, it is important who provides the feedback. To deliver value in
the end, the feedback should come from the people who will eventually use the product or service and
perhaps from those who maintain it. To measure this, we could count the average number of times
contact exists between the team and the end-users per user story.

First design
Now that a reference framework has been established the first design can be created. For this part the
Goal-Question-Metric (GQM) approach (Van Solingen & Berghout, 1999) is again used. As the goals have
been defined earlier (Table 9), questions and metrics can now be created.

Starting with the first goal of systems thinking, it is mainly defined meant as looking at the performance
of the system as a whole opposed to the separate parts. To understand and improve this performance, it
is important to know the current performance of the process for reference. Besides the current state, it

24
is important to know if a change in the process lead to an improvement in that performance. This leads
to the following two questions:

 What is the current performance of the process?


 Do changes in the process improve the performance of the entire process?

The second sub goal relates to the feedback loops within the system. Thus generally speaking how well
feedback can be given within the process and how well the process can deal with that feedback.
Structuring this in a similar fashion as the first sub goal, the first thing to know is the current state of
feedback loops. Besides that, while you can have a lot of feedback loops built into the process, this only
helps if the customer or end-user is satisfied with the feedback he can provide and if the system can
respond to the given feedback. This lead to the following questions:

 What is the current state of feedback loops within the process?


o Is the customer satisfied with the feedback that can be given?
o How well can the process respond to feedback?
 Do changes in the process improve the state of feedback loops within the process?

For the third sub goal of culture of learning and experimentation, the questions can again begin with the
current state of the culture. To answer this, research by López, Peón and Ordás (2004) can help by
defining four different aspects of organizational learning. The first construct relates to the acquisition of
knowledge, either through external sources or internal development. As a second construct they name
the distribution of knowledge followed by interpretation as a third construct. Interpretation describes
the way that people within the organization share and use the knowledge. The last construct they
describe is organizational memory, or the way the organization stores the knowledge for later use.
Besides learning, experimentation is there to further help improve the work and process and reward
trying new things (Kim, 2013).

Summarizing this, and again structuring like the previous two goals, this leads to the question what the
current state of learning and experimentation is like within the teams. This question can then be divided
further into three questions that could answer this, namely:

 What is the current state of learning and experimentation within the teams?
o Is knowledge being stored?
o Can knowledge be retrieved?
o Is the team improving their own process?
 Do changes in the process improve the state of learning and experimentation?

From these questions metrics can be defined that answer these questions and thus answer whether the
goal is reached. For the question of the first sub goal; do changes in the process improve the
performance of the entire process? the metrics should describe whether a difference exists between the
performance before and after a change. To describe the performance of a process Basili et al. (1994) use
the current average cycle time divided by a baseline average cycle time. To relate this to a change in the
process, and to the situation within Philips we could use the time from the moment a user story is

25
picked up until the moment it is being deployed and supported. Also the baseline would here be the
average before the change occurred. While this first metric only takes the speed into account, a second
metric could be used for quality. This is another factor that is important to take into account, as higher
quality leads to less rework, which in time leads to a better lead time. A simple metric that is already
being used within Philips could suffice here, by counting the average number of incidents after
deployment before and after the change in a similar way as the previous metric. A third metric to look at
would be the cost of a run through the process. This cost figure should be in there to account for the
fact, that if you would use double the resources for everything in the process, you would most likely also
see the cycle time being cut in half. Doing this however, will significantly increase or double the cost,
which in turn should not count as an improvement. An overview of the metrics related to the first goal is
shown in Table 10.

Table 10 GQM Metrics for system thinking


What is the current  𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑐𝑦𝑐𝑙𝑒 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑎 𝑢𝑠𝑒𝑟 𝑠𝑡𝑜𝑟𝑦
performance of the process?  𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑛𝑐𝑖𝑑𝑒𝑛𝑡𝑠 𝑎𝑓𝑡𝑒𝑟 𝑑𝑒𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡
 𝑇ℎ𝑒 𝑐𝑜𝑠𝑡 𝑜𝑓 𝑎 𝑢𝑠𝑒𝑟 𝑠𝑡𝑜𝑟𝑦 𝑔𝑜𝑖𝑛𝑔 𝑡ℎ𝑟𝑜𝑢𝑔ℎ 𝑡ℎ𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠

Do changes in the process 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑙𝑒𝑎𝑑 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑎 𝑢𝑠𝑒𝑟 𝑠𝑡𝑜𝑟𝑦 𝑎𝑓𝑡𝑒𝑟 𝑐ℎ𝑎𝑛𝑔𝑒
 ∗ 100%
improve the performance of 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑙𝑒𝑎𝑑 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑎 𝑢𝑠𝑒𝑟 𝑠𝑡𝑜𝑟𝑦 𝑏𝑒𝑓𝑜𝑟𝑒 𝑐ℎ𝑎𝑛𝑔𝑒
𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑛𝑐𝑖𝑑𝑒𝑛𝑡𝑠 𝑎𝑓𝑡𝑒𝑟 𝑑𝑒𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑎𝑓𝑡𝑒𝑟 𝑐ℎ𝑎𝑛𝑔𝑒
the entire process?  ∗ 100%
𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑛𝑐𝑖𝑑𝑒𝑛𝑡𝑠 𝑎𝑓𝑡𝑒𝑟 𝑑𝑒𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 𝑏𝑒𝑓𝑜𝑟𝑒 𝑐ℎ𝑎𝑛𝑔𝑒

To gain an understanding of the current state of feedback loops, the first thing that comes to mind is the
amount of feedback loops there are. While this sounds like a decent first metric, the issue here could be
that an arbitrary amount could be sufficient for a process of length x, but not for a process of length two
times x. To take the length of the process out of the metric a different metric can be used, which looks
at the average time between feedback moments. To use this metric, the start and end point of the
process should also count as a moment. Which means if no additional feedback moments are in place,
the time would be equal to the length of the process. Less abstractly, the time describes the time
between contact points with the customer. One problem with this would however be if the only
feedback moment is located at the end of the process, the average time would be same as if the
feedback moment would be right in the middle of the process. To cover this, an additional metric could
alleviate that problem, by keeping track of the maximum time within the process without feedback.
When this time would be very close to the average, we can see that the feedback moments are evenly
spread out over the process.

To answer if the customer is satisfied with the way he can provide feedback a qualitative metric is
proposed. This can be done by asking the customer whether or not he would have the next feedback
moment quicker than the time since the last feedback moment. This metric should atleast be asked the
first few times working with a particular customer, untill a clear understanding is reached between the
team and that customer.

26
To answer how well the system can respond to given feedback. A first suggestion would be to look at
the amount of work which has to be redone within the process. This can be quantified in the amount of
time spent from the moment of feedback until the process reaches the same point again. While this
might be slightly difficult to measure in practice, an easier metric could be to measure the total time
spent on rework during the process.

An overview of the second set of metrics relating to feedback loops is shown in Table 11.

Table 11 GQM Metrics for feedback loops


What is the current state  𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑖𝑚𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡𝑠
of feedback loops within  𝑀𝑎𝑥𝑖𝑚𝑢𝑚 𝑡𝑖𝑚𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡𝑠
the system?

Is the customer satisfied  𝑊𝑜𝑢𝑙𝑑 𝑡ℎ𝑒 𝑐𝑢𝑠𝑡𝑜𝑚𝑒𝑟 ℎ𝑎𝑣𝑒 𝑡ℎ𝑒 𝑛𝑒𝑥𝑡 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡 𝑞𝑢𝑖𝑐𝑘𝑒𝑟 𝑜𝑟 𝑙𝑎𝑡𝑒𝑟
with the feedback that 𝑡ℎ𝑎𝑛 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑖𝑛𝑐𝑒 𝑡ℎ𝑒 𝑙𝑎𝑠𝑡 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡?
can be given?
How well can the system  𝑇𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡 𝑢𝑛𝑡𝑖𝑙𝑙 𝑟𝑒𝑎𝑐ℎ𝑖𝑛𝑔 𝑡ℎ𝑒 𝑠𝑎𝑚𝑒 𝑝𝑜𝑖𝑛𝑡
respond to feedback?  𝑇𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑜𝑛 𝑟𝑒𝑤𝑜𝑟𝑘

To answer the questions about the current state of learning, the first and second metric could simply
state if new knowledge is actively being stored and if stored knowledge can be actively retrieved.
However to determine if knowledge is being shared as well as a mechanism in place to make sure that
knowledge is actually being stored a metric could state whether the team reflects on their work and
learnings after a project or sprint. Finally, a metric should state whether the team reflects on their way
of working, and thus takes time to improve their process or way of working. These metrics are shown
below in Table 12.

Table 12 GQM Metrics for culture of learning and experimentation


What is the current state of  𝐼𝑠 𝑛𝑒𝑤 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 𝑏𝑒𝑖𝑛𝑔 𝑠𝑡𝑜𝑟𝑒𝑑 𝑑𝑢𝑟𝑖𝑛𝑔 𝑡ℎ𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠?
learning and experimentation  𝐶𝑎𝑛 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠𝑙𝑦 𝑎𝑐𝑞𝑢𝑖𝑟𝑒𝑑 𝑘𝑛𝑜𝑤𝑙𝑒𝑑𝑔𝑒 𝑏𝑒 𝑟𝑒𝑡𝑟𝑖𝑒𝑣𝑒𝑑?
within the teams?  𝐷𝑜𝑒𝑠 𝑡ℎ𝑒 𝑡𝑒𝑎𝑚 𝑟𝑒𝑓𝑙𝑒𝑐𝑡 𝑜𝑛 𝑡ℎ𝑒𝑖𝑟 𝑤𝑜𝑟𝑘 𝑎𝑛𝑑 𝑙𝑒𝑎𝑟𝑛𝑖𝑛𝑔 𝑝𝑜𝑖𝑛𝑡𝑠
𝑎𝑓𝑡𝑒𝑟 𝑎 𝑝𝑟𝑜𝑗𝑒𝑐𝑡 𝑜𝑟 𝑠𝑝𝑟𝑖𝑛𝑡?
 𝐷𝑜𝑒𝑠 𝑡ℎ𝑒 𝑡𝑒𝑎𝑚 𝑟𝑒𝑓𝑙𝑒𝑐𝑡 𝑜𝑛 𝑡ℎ𝑒𝑖𝑟 𝑤𝑎𝑦 𝑜𝑓 𝑤𝑜𝑟𝑘𝑖𝑛𝑔
𝑎𝑓𝑡𝑒𝑟 𝑎 𝑝𝑟𝑜𝑗𝑒𝑐𝑡 𝑜𝑟 𝑠𝑝𝑟𝑖𝑛𝑡?

First feedback round and second design


Following the creation of the first design, two feedback sessions were performed. Both sessions were
done in the same setting, where the model was briefly explained per goal. After each goal the attendees
were asked what they thought of the goal, the questions and the metrics. After the last goal was
discussed, attendees had the chance to provide some additional feedback if they wished to do so. An
overview of the feedback can be found in Appendix E. The materials used can be found in Appendix F.
The materials included the process picture from Figure 7 to ensure those involved were all talking about
the same thing. This was especially helpful when talking about feedback loops. Also, to firstly explain

27
systems thinking being about the performance of the whole process, and not just the performance of
the separate teams. The first of the two sessions was done together with two people namely; Bas
Vermeer (Manager Operations, whom the assignment originated from) and Rita Conceicao (Global
Demand Manager). As the assignment originated from them, it would be useful to get their feedback.
Also, they would be able to adjust the direction if that turned out to be necessary. The second feedback
session was done with Laki Ahmed (IT Delivery Manager). Her position was closer to the teams that the
metrics are designed for. Therefore she could provide more specific feedback.

For the first goal, there were three main points of feedback. The first point considered the use of the
word ‘user story’, while it would be better to use a different unit of measurement. For the process, a
collection of user stories moves through the process simultaneously, except for the part where they are
developed. To measure this would not provide any additional information about the entire process. For
this reason, a new unit of measurement is chosen as the collection of these user stories, called a
‘feature’. The second point of feedback considered the specification of cost within the process, which
should be further defined. Considering the fact that this process contains knowledge work, thus no
tangible products, the cost of a feature would consist of the hours spent and the amount of people
working on it. That together is thus proposed as the cost measurement. The third point of feedback was
that by measuring the first question continuously, the second question would be irrelevant. As a
continuous measurement would show change in these metrics directly rather than as a change.
Therefore the second question has been removed. This led to the metrics in Table 13.

Table 13 GQM Metrics for system thinking


What is the current  𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑐𝑦𝑐𝑙𝑒 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑎 feature
performance of the  Average waiting time of a feature
process?  𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑛𝑐𝑖𝑑𝑒𝑛𝑡𝑠 as a result of the feature 𝑎𝑓𝑡𝑒𝑟 𝑑𝑒𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡

 𝑇h𝑒 𝑐𝑜𝑠𝑡 𝑜𝑓 𝑎 feature 𝑔𝑜𝑖𝑛𝑔 𝑡h𝑟𝑜𝑢𝑔h 𝑡h𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠


 Hours spent
 Number of people
 Number of features being worked on

For the second goal, the feedback was quite different over the two sessions. One session mainly agreed
with the proposed metrics and suggested some small changes in words and a few clearer definitions. For
the other session a different understanding of feedback loops led to some more feedback. The
understanding there meant more focus on feedback like how fast would changes in the system be seen
at the end of the process, rather than customer feedback. It was also suggested that I change feedback
moments to touch points for better understanding within Philips. This feedback led to a change in
questions, with more focus on the internal changes than the customer feedback. The changes can be
seen in Table 14.

28
Table 14 GQM Metrics for feedback loops
What is the current state  𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑖𝑚𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡𝑜𝑢𝑐ℎ𝑝𝑜𝑖𝑛𝑡𝑠
of feedback loops within  𝑀𝑎𝑥𝑖𝑚𝑢𝑚 𝑡𝑖𝑚𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡𝑜𝑢𝑐ℎ𝑝𝑜𝑖𝑛𝑡𝑠
the system?

How well can the system  𝑇𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡 𝑢𝑛𝑡𝑖𝑙𝑙 𝑟𝑒𝑎𝑐ℎ𝑖𝑛𝑔
respond to feedback? 𝑡ℎ𝑒 𝑠𝑎𝑚𝑒 𝑝𝑜𝑖𝑛𝑡
 𝑇𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑜𝑛 𝑟𝑒𝑤𝑜𝑟𝑘
How fast can the system  𝑇ℎ𝑒 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑖𝑚𝑒 𝑎 𝑐ℎ𝑎𝑛𝑔𝑒 𝑖𝑠 𝑠𝑒𝑒𝑛 𝑎𝑡 𝑡ℎ𝑒 𝑒𝑛𝑑 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠?
respond to changes in the  𝐶𝑦𝑐𝑙𝑒 𝑡𝑖𝑚𝑒 𝑑𝑖𝑣𝑖𝑑𝑒𝑑 𝑏𝑦 2
process? (How fast are the
changes visible in the
end?)

For the third goal, it was initially more difficult to find decent metrics. Therefore the feedback mainly
stated that the metrics were still mostly questions and should be further developed. Some feedback
included the addition of metrics related to the capabilities of the team members, and how well people
could perform the activities of other team members. Also a suggestion was made that the number of
value propositions should be counted. Here, a value proposition would mean a member making a
suggestion for change in the process or team with an estimated value that is gained by implementing
the change. Both these suggestions however, have their own problems. By just measuring the
capabilities, it would mean that you can get a culture of learning by simply hiring the people with the
most capabilities. While this is certainly important, it does nothing directly to measure your culture of
learning. The amount of value propositions would a nice metric to look at. However, you would also not
like your people to spend too much time on this, thus making it very difficult to say when this metric is
at a decent value.

However, some metrics are still proposed as further development was necessary. The focus for learning
lays on the time spent on improving daily work. This does not mean improving work done, and
improving it, but rather improving the way the teams perform their daily work. Thus measuring their
time spent on learning and partly in experimentation. For experimentation specifically some ideas were
suggested during the feedback sessions, which were added as well. The resulting questions and metrics
can be seen below in Table 15.

Table 15 GQM Metrics for culture of learning and experimentation


What is the current  𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑡𝑒𝑎𝑚𝑠 𝑠𝑝𝑒𝑛𝑑𝑠 𝑡𝑖𝑚𝑒 𝑖𝑚𝑝𝑟𝑜𝑣𝑖𝑛𝑔 𝑡ℎ𝑒𝑖𝑟 𝑑𝑎𝑖𝑙𝑦 𝑤𝑜𝑟𝑘?
state of learning  𝐻𝑜𝑤 𝑚𝑢𝑐ℎ 𝑡𝑖𝑚𝑒 𝑑𝑜 𝑡𝑒𝑎𝑚𝑠 𝑠𝑝𝑒𝑛𝑑 𝑜𝑛 𝑖𝑚𝑝𝑟𝑜𝑣𝑖𝑛𝑔 𝑡ℎ𝑒𝑖𝑟 𝑑𝑎𝑖𝑙𝑦 𝑤𝑜𝑟𝑘?
within the teams? (𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑡𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒)
 𝐼𝑠 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑎𝑑𝑒𝑞𝑢𝑎𝑡𝑒 𝑎𝑐𝑐𝑜𝑟𝑑𝑖𝑛𝑔 𝑡𝑜 𝑡ℎ𝑒 𝑡𝑒𝑎𝑚?

What is the current  𝐴𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑖𝑛𝑡𝑒𝑛𝑡𝑖𝑜𝑛𝑎𝑙 𝑓𝑎𝑢𝑙𝑡𝑠 𝑖𝑛𝑡𝑟𝑜𝑑𝑢𝑐𝑒𝑑 𝑝𝑒𝑟 𝑓𝑒𝑎𝑡𝑢𝑟𝑒?


state of  𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑓𝑎𝑢𝑙𝑡𝑠 𝑐𝑎𝑢𝑔ℎ𝑡 𝑏𝑒𝑓𝑜𝑟𝑒 𝐿𝑖𝑣𝑒?
experimentation
within the teams?

29
Second feedback round and third design
For the next feedback loop one single feedback session was held. This session was again held with Laki
Ahmed (IT Delivery Manager). The feedback in this session mainly consisted of small updates and
clarifications. Only for the third goal actual feedback was given to actually change things. This was
however more validation feedback than feedback to actually change the metrics. . An overview of the
feedback can be found in Appendix E. The materials used for this sessions can be seen in Appendix F.

For the first and second goal, this meant a small clarification in two specific terms. First, cycle time was
changed to lead time and Second, the cost of a feature was further specified by adding service costs. For
the metrics regarding experimentation, the feedback was that it might not be very useful to introduce
faults. Also, concerning there will be a pre-Live environment that is kept in the same state as Live for
testing purposes, the fault would be caught there anyway, and only cause extra work without much
gain. While this feedback is certainly valuable, it does not have to change the metric, but rather as a
validation step for the usefulness of the design.

The metrics model after these changes is depicted below in Table 16.

Table 16 DevOps Metrics Model


What is the current  𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑙𝑒𝑎𝑑 𝑡𝑖𝑚𝑒 𝑜𝑓 𝑎 feature
performance of the  Average waiting time of a feature
process?  𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑛𝑐𝑖𝑑𝑒𝑛𝑡𝑠 as a result of the feature 𝑎𝑓𝑡𝑒𝑟 𝑑𝑒𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡

 𝑇h𝑒 𝑐𝑜𝑠𝑡 𝑜𝑓 𝑎 feature 𝑔𝑜𝑖𝑛𝑔 𝑡h𝑟𝑜𝑢𝑔h 𝑡h𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠


 Hours spent
 Number of people
 Service costs
 Number of features being worked on

What is the current  𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑖𝑚𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡𝑜𝑢𝑐ℎ𝑝𝑜𝑖𝑛𝑡𝑠


state of feedback loops  𝑀𝑎𝑥𝑖𝑚𝑢𝑚 𝑡𝑖𝑚𝑒 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑡𝑜𝑢𝑐ℎ𝑝𝑜𝑖𝑛𝑡𝑠
within the system?  𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑝𝑝𝑟𝑜𝑣𝑎𝑙𝑠

How well can the  𝑇𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑓𝑒𝑒𝑑𝑏𝑎𝑐𝑘 𝑚𝑜𝑚𝑒𝑛𝑡 𝑢𝑛𝑡𝑖𝑙 𝑟𝑒𝑎𝑐ℎ𝑖𝑛𝑔
system respond to 𝑡ℎ𝑒 𝑠𝑎𝑚𝑒 𝑝𝑜𝑖𝑛𝑡
feedback?  𝑇𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑜𝑛 𝑟𝑒𝑤𝑜𝑟𝑘

How fast can the  𝑇ℎ𝑒 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑡𝑖𝑚𝑒 𝑎 𝑐ℎ𝑎𝑛𝑔𝑒 𝑖𝑠 𝑠𝑒𝑒𝑛 𝑎𝑡 𝑡ℎ𝑒 𝑒𝑛𝑑 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠?
system respond to  𝑙𝑒𝑎𝑑 𝑡𝑖𝑚𝑒 𝑑𝑖𝑣𝑖𝑑𝑒𝑑 𝑏𝑦 2
changes in the process?
(How fast are the
changes visible in the
end?)

30
What is the current state  𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑡𝑒𝑎𝑚𝑠 𝑠𝑝𝑒𝑛𝑑𝑠 𝑡𝑖𝑚𝑒 𝑖𝑚𝑝𝑟𝑜𝑣𝑖𝑛𝑔 𝑡ℎ𝑒𝑖𝑟 𝑑𝑎𝑖𝑙𝑦 𝑤𝑜𝑟𝑘?
of learning within the  𝐻𝑜𝑤 𝑚𝑢𝑐ℎ 𝑡𝑖𝑚𝑒 𝑑𝑜 𝑡𝑒𝑎𝑚𝑠 𝑠𝑝𝑒𝑛𝑑 𝑜𝑛 𝑖𝑚𝑝𝑟𝑜𝑣𝑖𝑛𝑔 𝑡ℎ𝑒𝑖𝑟 𝑑𝑎𝑖𝑙𝑦 𝑤𝑜𝑟𝑘?
teams? (𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑡𝑜𝑡𝑎𝑙 𝑡𝑖𝑚𝑒)
 𝐼𝑠 𝑡ℎ𝑒 𝑡𝑖𝑚𝑒 𝑠𝑝𝑒𝑛𝑡 𝑎𝑑𝑒𝑞𝑢𝑎𝑡𝑒 𝑎𝑐𝑐𝑜𝑟𝑑𝑖𝑛𝑔 𝑡𝑜 𝑡ℎ𝑒 𝑡𝑒𝑎𝑚?

What is the current state  𝐴𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑖𝑛𝑡𝑒𝑛𝑡𝑖𝑜𝑛𝑎𝑙 𝑓𝑎𝑢𝑙𝑡𝑠 𝑖𝑛𝑡𝑟𝑜𝑑𝑢𝑐𝑒𝑑 𝑝𝑒𝑟 𝑓𝑒𝑎𝑡𝑢𝑟𝑒?
of experimentation  𝑃𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑜𝑓 𝑓𝑎𝑢𝑙𝑡𝑠 𝑐𝑎𝑢𝑔ℎ𝑡 𝑏𝑒𝑓𝑜𝑟𝑒 𝐿𝑖𝑣𝑒?
within the teams?

As so few changes were actually made, this is the last feedback cycle performed. Also, in this last cycle,
part of the feedback belongs more closely to validation than feedback. As during the sessions an
agreement was reached on each part except for the culture of experimentation. This suggests the
proposed design consists of metrics that would be useful within Philips IT to measure the progress on
DevOps. For the state of experimentation, the feedback however was that it would not be very useful to
actually do this experimentation. For this reason, it is still kept in the model, as it is a part of what
DevOps is about. However, for the usability of the metrics it could be excluded by Philips if there is no
perceived value. The reason for this is that the faults would only cause extra work, and not actually
create any additional value.

31
7. Discussion

In this chapter some theoretical and practical points are discussed. This starts with a small discussion on
the criticism in literature regarding the GQM approach. Following that, the creation of process models
as a start to design metrics is discussed. Finally the final metrics model is discussed in contrast to the
metrics found earlier in the literature.

While GQM is an approach that has quite some positive response in literature, criticism states that the
outcome is rather unpredictable as it is still possible to come to many different metrics that describe the
goal. While this is true, this research shows that by performing a few feedback cycles you can adapt
those first metrics into metrics that can be validated and used within the environment.

As suggested by Pfleeger (1993), this research also started to design metrics by first describing the
process. As someone not working daily in these processes the creation of these process models
provided a much better understanding of what was going on. However, besides the value of creating an
understanding what the people in the process are doing, the models were hardly used during the design
of the metrics model. The main use came from illustrating where the metrics could be applied and to
clarify concepts like feedback loops and systems thinking. The amount of detail the processes were
modelled in, was probably not needed for this particular purpose. A quick and small overview of the
process is most likely enough to use as illustration while designing the metrics.

While not all metrics in the final model can be directly found within the literature research, some clear
similarities can be found. For the first goal of systems thinking parallels are found in the average lead
time and the amount of features being worked on simultaneously (WIP). Both of these were also
included in the research by Duque & Cadavid (2007). The other metrics in the model were not directly
found. However, these metrics represent costs and quality, thus might not have been included in the
found literature because they are not specific to the domain of lean manufacturing or scrum. For the
second goal the parallels are more hidden, but they are most certainly present. While different, the time
spent on rework is mentioned again in literature as the percentage of units sent for rework (Duque &
Cadavid, 2007). Other than that, the metrics found, like approvals, are more closely related to software
development and were less present in manufacturing. For the third goal, a culture or learning and
experimentation, the metrics turned out quite different than what was previously found in literature
(López, Peón, & Ordás, 2004) as during the first feedback sessions, this seemed to be too abstract.
Therefore some simpler and more direct metrics were chosen.

Looking at the metrics model from the literature, the lean manufacturing metrics turned out to be quite
helpful in especially the first goal. The metrics found for scrum development however have not been
used in the designed metrics. The reason for this is most likely the focus of these metrics. While the lean
manufacturing metrics focused on the entire process, similar to the focus of the designed metrics
model, the scrum metrics focused more on teams working within this entire process. Therefore these
were too specific to answer the goals.

32
8. Conclusion

This research has provided a metrics model to measure the progress of DevOps within Philips IT. While
the validation currently showed low interest for the metrics related to culture of experimentation, this
could become more useful when DevOps is further developed within Philips. As more automation would
be implemented and more time becomes available to actually try this experimentation.

While the designed metrics model could not yet be implemented within Philips to gather some
measurements, this research has shown that within a few feedback loops a clear metrics model can be
designed using the GQM approach.

33
References

Ackoff, R. L. (1971). Towards a system of systems concepts. Management science, 17(11), 661-671.

Anderson, D. J. (2010). Kanban: Successful evolutionary change for your technology business. Blue Hole
Press.

Basili, V. R., Caldiera, G., & Rombach, H. D. (1994). Goal Question Metric Paradigm. Encyclopedia of
Software Engineering, 1, 528-532.

Basili, V. R., Caldiera, G., & Rombach, H. D. (1994). The goal question metric approach. Retrieved 2014,
from http://www.cs.umd.edu/~mvz/handouts/gqm.pdf

Beck, K., & Andres, C. (2000). Extreme programming explained: Embrace change. Addison-Wesley.

Beck, K., Beedle, M., van Bennekum, A., Cockburn, A., Cunningham, W., Fowler, M., et al. (2001).
Retrieved 2014, from The Agile Manifesto: http://agilemanifesto.org/

Conboy, K., & Fitzgerald, B. (2004). Toward a conceptual framework of agile methods: A study of agility
in different disciplines. 2004 ACM workshop on Interdisciplinary software engineering research,
(pp. 37-44). Newport Beach.

DevOps. (2014, January 7). Retrieved January 14, 2014, from Wikipedia:
http://en.wikipedia.org/wiki/DevOps

Downey, S., & Sutherland, J. (2013). Scrum metrics for hyperproductive teams: How they fly like fighter
aircraft. System Sciences (HICSS), 2013 46th Hawaii International Conference on System Sciences,
(pp. 4870-4878). Hawaii.

Duque, D. F., & Cadavid, L. R. (2007). Lean manufactoring measurement: The relationship between lean
activities and lean metrics. Estudios Gerenciales, 23(105), 69-83.

Erickson, J., Lyytinen, K., & Siau, K. (2005). Agile modeling, agile software development, and extreme
programming: The state of research. Journal of database management, 16(4).

Goldratt, E. M., & Cox, J. (1985). The goal: A process of ongoing improvement. North River Press.

Ikonen, M., Kettunen, P., Oza, N., & Abrahamsson, P. (2010). Exploring the sources of waste in kanban
software development projects. 36th EUROMICRO conference on software Engineering and
advanced applications (SEAA).

ITIL 2011 - The big picture. (2011). Retrieved 1 20, 2014, from
http://cfnpeople.com/downloads/itil_poster_the_big_picture_cfn_people.pdf

Jailia, K., Sujata, Jailia, M., & Agarwal, M. (2011). Lean software development ("As a survival tool in
recession"). International journal of software engineering and its applications, 5(3).

34
Karlsson, C., & Ahlstrom, P. (1996). Assessing changes towards Lean Production. International Journal of
Operations & Production Management, 16(2), 24-41.

Kim, G. (2013, January 22). DevOps distilled, Part 1: The three underlying principles. Retrieved January
2014, from IBM Developerworks: http://www.ibm.com/developerworks/library/se-
devops/part1/index.html

Kim, G., Behr, K., & Spafford, G. (2013). The phoenix project: A novel about IT, DevOps, and helping your
business win. IT Revolution Press.

Leffinwell, LLC. (2014). Scaled Agile Framework. Retrieved April 24, 2014, from
http://scaledagileframework.com/

López, S. P., Peón, J. M., & Ordás, C. J. (2004). Managing knowledge: the link between culture and
organizational learning. Journal of knowledge management, 8(6), 93-104.

Martinez, A., & Perez, M. (2001). Lean indicators and manufacturing strategies. International Journal of
Operations & Production Management, 21(11), 1433-1451.

McNiff, J. (2013). Action research; Principles and practice. London & New York: Routledge.

Nissen, Christian Feldbech; CFN People. (2010). IT Service Management.

Petersen, K., & Wohlin, C. (2010). Software process improvement through the lean measurement (SPI-
LEAM) method. The journal of systems and software 83, 1275-1287.

Petersen, K., & Wohlin, C. (2011). Measuring the flow in lean software development. Software: Practive
and Experience(41), 975-996.

Pfleeger, S. L. (1993). Lessons learned in building a corporate metrics program. IEEE Software, 67-74.

Philips IT. (2013). IT Transformation: Creating the Philips Integrated Landscape.

Poppendieck, M., & Poppendieck, T. (2003). Lean software development: An agile toolkit for software
development managers. Boston: Addison-Wesley.

Rally Software Development Corp. (2014). Retrieved 4 24, 2014, from Rally Software:
http://www.rallydev.com

Royce, W. W. (1970). Managing the development of large software systems. IEEE Wescon, (pp. 1-9).

Schwaber, K., & Beedle, M. (2001). Agile development with scrum. Prentice Hall.

Stapleton, J. (2003). DSDM: Business Focused Development, second edition. Pearson Education.

Van Solingen, R., & Berghout, E. (1999). The goal question metric method: a practical guide for quality
improvement of software development. McGraw-Hill Inc.

35
Womack, J. P., Jones, D. T., & Roos, D. (1990). The machine that changed the world: the story of lean
production - Toyota's secret weapon in the global car wars that is revolutionizing world industry.

Yin, R. K. (1989). Case study research design and methods. Newbury Park: Sage Publications.

Young, K. W., Muchlhaeusser, R., Pigging, R. S., & Rachitrangsan, P. (2001). Agile control systems. Journal
of automobile engineering.

Zhang, Z., & Sharifi, H. (2000). A methodology for achieving agility in manufacturing organisations.
International journal of operations & production management, 20(4), 496-512.

36
Appendix A
Original project description

Design DevOps for Philips IT

Are you interested to implement DevOps in Philips IT through designing the new working processes
across IT Development and Operations based on Logistical and Lean principles?

Background

Philips IT is a fully centralized IT organization servicing the 3 sectors of Philips (Healthcare, Lighting and
Consumer Lifestyle). Within the IT organization IT projects are executed in IT Delivery, where daily
Operations is executed within Philips IT Infrastructure & Operations (I&O). Over the last 2 years Philips
IT made a significant change towards Agile/SCRUM project management. At this moment, Philips is one
of the leaders in the industry with more than 100 SCRUM project teams delivering working software
every 2 weeks. We do this in full multidisciplinary teams together with our partners in an completely
virtual way across the globe (teams working together between Europe, US, India on the same software
products for instance).

Philips IT Operations (I&O) is responsible for the day to day operations. With every release of new
functionality IT Delivery hands over to IT I&O for day to day operations. In IT I&O we work
according to ITIL, which is the Industry best practice operations process framework. ITIL is sometimes
conflicting with the Agile/SCRUM mindset. Where Agile/SCRUM demands faster software releases and
shorter cycle times, from an operation perspective this may lead to additional risks to system stability,
security and performance. An explicit risk is that scrum teams do not spend sufficient time on the full
test cycle (regression testing, stress testing, performance testing) leading to production issues and
outages.

To overcome this IT Delivery and IT I&O are working together to introduce DevOps into the IT
organization. DevOps is a new way of working between Delivery and Operations based on 3 principles:
Fast feedback loops through multiple releases per week and even per day creating Robust solutions;
Perfect deployment through fully automated testing and deployment procedures for UAT, regression
and stress testing and release mgt; and Rebuild instead of Repair. A full description on DevOps thinking
is described in the Phoenix Project (Gene Kim, Kevin Behr and George Spafford).

To implement DevOps properly within Philips IT we are looking for students that want to support us in
implementing this new way of working in Philips.
Assignment

We have an open position to support this change in Philips around the design and implementation of
DevOps. How should delivery and operations change their Agile/Scrum and ITIL work processes to
enable a full DevOps way of working within IT? How should they adopt principles from Logistics and
Lean to coordinate the workload and ensure high quality delivery of new software?

37
The assignment is to Design the new working processes between Delivery and Operations based on
Logistical and Lean operating principles to enable daily releases of new functionality in production.

Candidate Profile

For this assignment we are looking for a student on a university level with a background in Industrial
Engineering, Logistics and/or Information Technology with a clear interest in Logistical design and using
Lean principles on work processes. Student should be in his/her final year of a Masters Degree.

Location

Location of the assignment is Eindhoven, High Tech Campus building 42/43.

Further info

For further information please contact Bas Vermeer (Bas.Vermeer@Philips.com).

38
Appendix B
Philips Integrated Landscape: Information Management

39
40
41
42
43
Appendix C
VIPP
VIPP: improved management decision support system

44
45
46
47
48
Appendix D
Lean measurements (Duque & Cadavid, 2007)

Dimension Metrics
Elimination of waste  Work in Progress
 Setup time
 Machine downtime
 Transportation
 Space Utilization
Continuous improvement  Number of suggestions per employee per
year
 Number of suggestions that get
implemented
 Scrap %
 Rework %
Continuous flow and Pull-driven systems  Lot sizes
 Order flow time
 Order lead time
 Pulling processes
 Pull Value %
Multifunctional teams  Autonomous control
 Workteam Task Content
 Cross training
 Number of employees capable of
assignment rotation
Information Systems  Frequency with which information is given
to employees
 Percentage of procedures that are
documented in the company
 Frequency with which the line or cell
progress boards are updated

49
Appendix E
Feedback session 1 (Round 1)
Date: July 16th 2014
Participants: Bas Vermeer (Manager Operations) & Rita Conceicao (Global Demand Manager)

Goal or Metric Remarks Changes


Systems thinking Made specific use of the process picture to
ensure we were all talking about the
performance of the entire process opposed
to separate parts within the process
The cost of a user story Should be further specified, what does a Added: Hours spent & number
going through the cost consist of of people
process
Feedback Loops More focus on how the system responds to Added: How fast can the
feedback and not so much on customer system respond to changes in
feedback the process? (How fast are the
changes visible in the end?)
Removed: Is the customer
satisfied with the feedback
that can be given?
Culture of Learning Metrics were still pretty much questions No direct changes during the
and not really metrics. Suggestions were to session
measure the capabilities of team members
(capability matrix)
Culture of Some suggestions were made regarding Added:
Experimentation introducing faults into the system and  Amount of intentional
measuring those. faults introduced
 Percentage of faults
caught before Live

50
Feedback session 2 (Round 1)
Date: July 21st 2014
Participants: Laki Ahmed (IT Delivery Manager)

Goal or Metric Remarks Changes


Systems thinking Made specific use of the process picture to Change user story to feature
ensure we were all talking about the (collection of user stories)
performance of the entire process opposed
to separate parts within the process.
User story as an object was deemed too
small.
What is the current  Cost should be further specified Added:
performance of the (same as session 1).  Average waiting time
process?  Add a measure to see how much of a feature
more can be improved.  Number of features
 Work in progress or done being worked on
simultaneously could also be
important
Do changes in the Unnecessary to measure. By measuring the Remove question
process improve the performance continuously, changes can be
performance of the seen much faster.
entire process?

Feedback Loops  Concept explained along the Change “feedback moment” to


process model “Touchpoint”
 Internal use of the term
“Touchpoint” instead of feedback
moment
Culture of Learning Metrics were still pretty much questions No direct changes during the
and not really metrics. Suggestions were to session
measure the amount of value propositions
made by team members
Culture of Showed interest in the concept but no
Experimentation suggestions were made

51
Feedback session 3 (Round 2)
Date: September 17th 2014
Participants: Laki Ahmed (IT Delivery Manager)

Goal or Metric Remarks Changes


Systems thinking Cycle time was incorrectly used as a term. Change cycle time to lead time
Should change to lead time
The cost of a user story Should be further specified, should include Added: Service costs
going through the service costs
process
Culture of  Live should be pre-Live as an extra Change Live to pre-Live
Experimentation environment will exist as a copy of
Live, but for testing purposes.
 Did not perceive much added value
in introducing faults, especially
because of the pre-Live
environment

52
Appendix F
1st Feedback round

53
2nd Feedback round

54

You might also like