Queuing Theory

Queuing Theory and
Measurement of Effectiveness
Simulation & Modeling

Kardi Teknomo, PhD
Kardi Teknomo, PhD
Overview
Queuing Theory
Arrival distribution
Service time distribution
Kendal notation
Measurement of effectiveness
M/M/1
M/M/s
Queuing Optimization
Optional: Queue with Limits
Kardi Teknomo, PhD
Objectives
By the end of this lesson you should be able to
Compute Measurement of Effectiveness (MOE) of

simple queuing with multi channel
Find cost effective queuing channel
Kardi Teknomo, PhD
What is queuing theory?
Queue = waiting line

Q
Queuing
i theory
th
= the
th body
b d off knowledge
k
l d
dealing with waiting lines
Kardi Teknomo, PhD
Queuing Model
Kardi Teknomo, PhD
Queueing systems are pervasive
Banks
Traffic
Hospitals
Go e nment offices
Government
Supermarkets
R t
Restaurants
t
Documents in office
Files
Products queue for Machine
Kardi Teknomo, PhD
Queueing System
Input:
Arrival Rate & Distribution

Service rate & Distribution
Configuration
g
(e.g.
( g Parallel,, serial))
Waiting Capacity (Max queue length)
Population (finite or infinite)
Service Discipline
Output:
Measurement of Effectiveness (MOE) or operating

characteristics (performance measures)
Kardi Teknomo, PhD
Arrival Distribution
Most common type
Markovian (= Poisson Process)

Deterministic (e.g. constant or linear by time)
If the number of arrival in a given period of

time occurs randomlyy and independently
p
y from
other arrivals and follows a Poisson
distribution with mean customers/minute,
th th
then
the interarrival
i t
i l time
ti
(h d
(headway)
)
distribution follow an exponential probability
distribution with mean 1/ minutes
Kardi Teknomo, PhD
Poisson Probablity Distribution
Distribution of arrivals
Probability of x arrivals in a specific time
period:
P ( x) =
N ti
Notice:
PMF
x e
x!
, x = 0,1, 2,"
Kardi Teknomo, PhD
Practice-1: Poisson Distribution
Teknomos Burger has analyzed the data on

customer arrival & found that the disribution is
Poisson and the mean arrival rate is 45
customers/60 minutes. Compute
p
the probability
p
y
of x customer arrive during a 1 minute period
for x = 0, 1, 2, 3, 4, >=5
Kardi Teknomo, PhD
Answer Practice 1
= 45/60 minutes = 0.75 customers/minute

75/0! = e-0
0.75
75 = 0.4724
P(0)=(0 75)0e-00.75
P(0)=(0.75)
0 4724
P(1)=(0.75)1e-0.75/1! = 0.75 e-0.75 = 0.3543
Etc
Number of Arrivals
Probability
0.4724
0.3543
0.1329
0.0332
0.0062
5 or more
0 0010
0.0010
Kardi Teknomo, PhD
Service distribution
Most common
Markovian (Poisson Process)

Deterministic
General ((e.g.
g Unspecified/Normal
p
/
distribution))
Service rate = =average number of

customers that can be served p
per time period
p
(Poisson Dist.)
Average service time per customer = 1/ time
period (Exponential distributed)
Kardi Teknomo, PhD
Exponential distribution
Probability that the service time will be

less than or equal to a time length t is
P ( service time t ) = 1 e
e = 2.71828
Notice: CDF
Kardi Teknomo, PhD
Example (Practice 2)
Teknomos Burger has studied that order filling

process for single food server can process on
average 60 customers orders per hour.
Compute
p
the mean service rate per
p minute and
probaility that an order can be processed in
0.5 minutes or less,

1 minute or less,
2 minutes or less.
Kardi Teknomo, PhD
Answer Practice 2
Mean service rate = 60 customers/60

minutes = 1 customer/minute
P(service time 0.5 min) = 1 - e-1(0.5) = 1 - 0.6065 =

0.3935
P(service time 1.0 min) = 1 - e-1(1.0) = 1 - 0.3679 =
0.6321
1(2 0) = 1 - 0.1353
P(service time 2.0
2 0 min) = 1 - e-1(2.0)
0 1353 =
0.8647
Kardi Teknomo, PhD
Queueing Discipline
FIFO = First In First Out = FCFS = First come

first served
Others:
LIFO = Last In First Out

Balk = don
dontt join queue upon arrival
Renege = leave queue before being served
Jockey = swicth from one queue to another
Loop = join the same queue after served
Priority = certain type of customers will be served
fi t
first
Kardi Teknomo, PhD
Measurement of Effectiveness
U= Utilization factor = percentage of the time that all

servers are busyy
P0 = probability that there are no customers in the
system
Pn = probability
b bili that
h there
h
are n customers in
i the
h
system
Pw = probability that an arriving customer has to wait
for service
Kardi Teknomo, PhD
MOE (contd)
Wq = average time a customer spends in waiting line

waiting
g for service
W = average time a customer spends in the system (in
waiting line and being served)
Lq = average number
b off customer iin waiting
i i line
li for
f
service
L = average number of customer in the system (in
waiting line and being served)
Kardi Teknomo, PhD
Steady State & Transient
In the beginning the queuing system starts in

empty and idle condition then gradually go
through transient periods before the activities
reach steadyy state level
Queuing MOE formulas are derived from
Markov Chain to describe only steady state
(normal) operation.
Simulation will also show operation in both
transient and steady state condition
Kardi Teknomo, PhD
Kendall Notation
Kendal suggested a notation that is helpful in

classifying wide variety of different waiting line
A/B/k
A = prob.
prob distribution of arrivals
B = prob. distribution of service time
k = number of channels
Commonly used letter in position A or B are:
M = Poisson process (Poisson dist. for arrivals and

exponential dist. for service time)
D = Deterministic or constant
G=G
Generall prob.
b di
dist.
t With known
k
mean and
d
variance
Kardi Teknomo, PhD
MOE of M/M/1
Utilization factor
Prob. of n units in system
Prob. Empty system
U = = <1
Pn = n P0
P0 = 1 = 1
If utilization factor is larger to equal to 1

1, the waiting
line will continue to grow without limit
Kardi Teknomo, PhD
MOE of M/M/1 (contd)

1
Avg Waiting in Queue

Avg.
Wq = W
Avg. Waiting in System
1
W=

Avg. Queue Length
2
Lq = Wq =
( )
Avg. System Length
L = W = Lq +
Kardi Teknomo, PhD
Practice 3: M/M/1
Poisson Process (M/M/1)

= 0.75
0 75 customers/minute
= 1 customer/minute
Compute MOE of Teknomo
Teknomoss Burger:
U, P0, Lq, L, W, Wq
Pn where n = 0 to (7 or more)
Kardi Teknomo, PhD
Answer Practice 3
MOE of M/M/1: = 0.75, = 1

0.75
0 75
U == =
= 0.75 < 1 (OK )
1
2
0 752
0.75
Lq =
=
= 2.25 customers
( ) 1(1 0.75 )
L = Lq + = 2.25 + 0.75 = 3 customers

1
1
W=
=
= 4 minutes
i t
1 0.75
1
1
Wq = W = 4 = 3 minutes
1
Kardi Teknomo, PhD
Answer Practice 3 (contd)
0.75
= 0.25
P0 = 1 = 1 = 1
1
Pn = P0 = ( 0.75 ) 0.25
n
Numberofcustomers
0
1
2
3
4
5
6
7ormore
Probability
25.00%
18.75%
14.06%
10 55%
10.55%
7.91%
5.93%
4.45%
13.35%
Sum
25.00%
43.75%
57.81%
68 36%
68.36%
76.27%
82.20%
86.65%
100.00%
Kardi Teknomo, PhD
Effect of Changing units
Instead of using = 0.75

customers/minute and = 1
customer/minute, you might want to use
= 45 customers/hour and = 60
customers/hour
Will the MOEs change because we
change the units?
Kardi Teknomo, PhD
Effect of Changing units (contd)
Kardi Teknomo, PhD
Improving Waiting Line Operation
To reduce waiting time, we must focus

on service improvements:
1.
2
2.
Increase the mean sevice rate by making a

creative design change or by using new
technology
Add service channels so that more
customers can be served simultaneously
Kardi Teknomo, PhD
Practice 4: M/M/1
improved service
Teknomo Burger decide to employ an order

filler who will assist the order taker
taker. With this
design the management estimates the mean
service rate can increase from 60 customers
per hour into 75 customers per hour.
Compute the MOEs and compare the results to
practice 3
Kardi Teknomo, PhD
Answer Practice 4
MOE of M/M/1: = 0.75, = 1.25

0.75
0 75
U == =
= 0.60 < 1 (OK )
1.25
2
0 752
0.75
Lq =
=
= 0.9 customers
( ) 1.25 (1.25 0.75 )
L = Lq + = 0.90 + 0.60 = 1.5 customers
1
1
W=
=
= 2 minutes
i t
1.25 0.75
1
1
Wq = W = 2
= 1.2 minutes
1.25
Kardi Teknomo, PhD
Answer Practice 4 (contd)
0.75
P0 = 1 = 1 = 1
= 0.40
1.25
Pn = P0 = ( 0.60 ) 0.40
n
Numberofcustomers
0
1
2
3
4
5
6
7ormore
Probability
40.00%
24.00%
14.40%
8 64%
8.64%
5.18%
3.11%
1.87%
2.80%
Sum
40.00%
64.00%
78.40%
87 04%
87.04%
92.22%
95.33%
97.20%
100.00%
Kardi Teknomo, PhD
MOE of M/M/s
Utilization factor
U = 1s

s
+
P0 =
n =0 n ! s ! s
s 1
Prob. Empty system
P0 s +1
Avg. Queue Length
Lq =
Avg. System Length
L = Lq +
( s 1)!( s )
Kardi Teknomo, PhD
MOE of M/M/s (contd)
Avg. Waiting in Queue
Prob. Of Waiting
Wq =
Lq
W = Wq +
s
Pw =
P0
s ! ( s )
n
P0

n!
Pn = n
P
s ! s n s 0
for n s
for n > s
Kardi Teknomo, PhD
Practice 5: multi servers
If the management of Teknomo burger want to

use 2nd order processing station so that
customers can bee served simultaneously, what
are the MOEs if
Arrival rate = 0.75 customers/min
Service rate = 1 customer/min
/
Kardi Teknomo, PhD
Answer practice-5
M/M/2: = 0.75, = 1.0

Build Spreadsheet
Compute P0 first
Then compute other MOEs
Kardi Teknomo, PhD
Economic Analysis of Queue
Mostly based on subjective evaluation of operating

characteristics
More scientific way is to use Total Cost model
TC = Cw L + Cs k
TC = total cost per time period

Cw =waiting cost per time period per each unit
L = average number of unit in the system
Cs = service cost per time period for each channel
k = number of channels
Kardi Teknomo, PhD
General shape of total cost model
Kardi Teknomo, PhD
Practice 6: economic analysis
Use spreadsheet of practice 5 to analyze total

cost model of Teknomos
Teknomo s burger given that
= 0.75, = 1.0
Server wages,
g , benefit and other direct costs is
P700/hour
Opportunty cost of losing a customer is P1000/hour
for customer waiting time
What is the most economical number of

channels?
Kardi Teknomo, PhD
Answer Practice 6
Input: Number of channel k=1,2,...,10

Obtain: Average system length, L
C
Compute
& chart:
h
Service
S i Cost
C
and
d Waiting
W i i Cost,
C
Total
T l cost
Optimum number of channels = 2
Kardi Teknomo, PhD
Queue with Limits
Kendal notation: A/B/k/C/D

C = Capacity
C
it off the
th queue (finite
(fi it total
t t l
seats or limited space)
D = Domain or population that may seek
service is finited (from which the arrivals
are derived)
Kardi Teknomo, PhD
MOE of M/M/s/K
U = 1s ( L Lq )
Utilization factor
Prob. Empty system P0 = 1 +

+
n =1 n ! s !
Avg. Queue Length
Lq =
P0 s
s !(1 )
(1
Avg System Length

Avg.
K s
( K s )
K s

n = s +1 s
K
(1 ) ) ,
s 1
L = nPn + Lq + s 1 Pn
n =0
n =0
s 1
ns
Kardi Teknomo, PhD
MOE of M/M/s/K (contd)

Lq
Wq =
L
W=
(1 PK )
n
P0
n!
n
Pn = n s P0
s !s
0
(1 PK )
for n = 1, 2," , s
for n = s + 1, s + 2," , K
for n > K
Kardi Teknomo, PhD
Practice 7
Compute MOE of queue with Poisson process

for both arrival rate and service rate with finite
queue K=5
Arrival rate = 5, service rate = 7
Number of server: 1, 2
Kardi Teknomo, PhD
MOE of M/M/s//N
Prob. Empty system
N !
N !
+
P0 =
ns
n =0 ( N n )!n !
N
n
!
s
!
s
(
)
n=s
s 1
Avg. Queue Length

N
Lq = ( n s ) Pn
n=s
Avg System Length

Avg.
s 1
L = nPn + Lq + s 1 Pn
n =0
n =0
s 1
Kardi Teknomo, PhD
MOE of M/M/s//N (contd)

Lq
Wq =
L
W=
( N L)
N ! n P0
( N n )!n !
N ! n P
0
Pn =
ns
!
!
N
n
s
s
(
)
( N L)
for 0 < n s
for s < n N
for n > N
Kardi Teknomo, PhD
Practice 8
Compute MOE of queue with Poisson process

for both arrival rate and service rate with finite
population N=10
Arrival rate = 0.01, service rate = 0.125
Number of server: 1, 2

Queuing Theory

Uploaded by

Copyright:

Available Formats

Queuing Theory

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Queuing Theory

Uploaded by

Copyright:

Available Formats

Queuing Theory and

Simulation & Modeling

Kardi Teknomo, PhD

Kardi Teknomo, PhD

By the end of this lesson you should be able to

Compute Measurement of Effectiveness (MOE) of

Kardi Teknomo, PhD

What is queuing theory?

Queue = waiting line

Kardi Teknomo, PhD

Kardi Teknomo, PhD

Queueing systems are pervasive

Kardi Teknomo, PhD

Arrival Rate & Distribution

Measurement of Effectiveness (MOE) or operating

Kardi Teknomo, PhD

Most common type

Markovian (= Poisson Process)

If the number of arrival in a given period of

Kardi Teknomo, PhD

Poisson Probablity Distribution

Kardi Teknomo, PhD

Practice-1: Poisson Distribution

Teknomos Burger has analyzed the data on

Kardi Teknomo, PhD

= 45/60 minutes = 0.75 customers/minute

Kardi Teknomo, PhD

Markovian (Poisson Process)

Service rate = =average number of

Kardi Teknomo, PhD

Probability that the service time will be

Kardi Teknomo, PhD

Teknomos Burger has studied that order filling

0.5 minutes or less,

Kardi Teknomo, PhD

Mean service rate = 60 customers/60

P(service time 0.5 min) = 1 - e-1(0.5) = 1 - 0.6065 =

Kardi Teknomo, PhD

FIFO = First In First Out = FCFS = First come

LIFO = Last In First Out

Kardi Teknomo, PhD

U= Utilization factor = percentage of the time that all

Kardi Teknomo, PhD

Wq = average time a customer spends in waiting line

Kardi Teknomo, PhD

Steady State & Transient

In the beginning the queuing system starts in

Kardi Teknomo, PhD

Kendal suggested a notation that is helpful in

Commonly used letter in position A or B are:

M = Poisson process (Poisson dist. for arrivals and

Kardi Teknomo, PhD

Prob. of n units in system

Prob. Empty system

If utilization factor is larger to equal to 1

Kardi Teknomo, PhD

MOE of M/M/1 (contd)

Avg Waiting in Queue

Avg. Waiting in System