Chapter 6

Queueing Models
Banks, Carson, Nelson & Nicol
Discrete-Event System Simulation

Purpose
Simulation is often used in the analysis of queueing models.
A simple but typical queueing model:
Queueing models provide the analyst with a powerful tool for
designing and evaluating the performance of queueing
systems.
Typical measures of system performance:
Server utilization, length of waiting lines, and delays of customers
For relatively simple systems, compute mathematically
For realistic models of complex systems, simulation is usually
required.

Outline
Discuss some well-known models (not development
of queueing theories):
General characteristics of queues,
Meanings and relationships of important performance
measures,
Estimation of mean measures of performance.
Effect of varying input parameters,
Mathematical solution of some basic queueing models.

Characteristics of Queueing Systems

Key elements of queueing systems:
Customer: refers to anything that arrives at a facility and requires
service, e.g., people, machines, trucks, emails.
Server: refers to any resource that provides the requested
service, e.g., repairpersons, retrieval machines, runways at
airport.

Calling Population
[Characteristics of Queueing System]
Calling population: the population of potential customers,
may be assumed to be finite or infinite.
Finite population model: if arrival rate depends on the number of
customers being served and waiting, e.g., model of one corporate
jet, if it is being repaired, the repair arrival rate becomes zero.
Infinite population model: if arrival rate is not affected by the
number of customers being served and waiting, e.g., systems
with large population of potential customers.

System Capacity
[Characteristics of Queueing System]
System Capacity: a limit on the number of customers
that may be in the waiting line or system.
Limited capacity, e.g., an automatic car wash only has room for
10 cars to wait in line to enter the mechanism.
Unlimited capacity, e.g., concert ticket sales with no limit on the
number of people allowed to wait to purchase tickets.

Arrival Process
[Characteristics of Queueing System]
For infinite-population models:
In terms of interarrival times of successive customers.
Random arrivals: interarrival times usually characterized by a
probability distribution.
Most important model: Poisson arrival process (with rate ), where
A
n
represents the interarrival time between customer n-1 and
customer n, and is exponentially distributed (with mean 1/).
Scheduled arrivals: interarrival times can be constant or constant
plus or minus a small random amount to represent early or late
arrivals.
e.g., patients to a physician or scheduled airline flight arrivals to an
airport.
At least one customer is assumed to always be present, so the
server is never idle, e.g., sufficient raw material for a machine.

Arrival Process
[Characteristics of Queueing System]
For finite-population models:
Customer is pending when the customer is outside the queueing
system, e.g., machine-repair problem: a machine is pending
when it is operating, it becomes not pending the instant it
demands service form the repairman.
Runtime of a customer is the length of time from departure from
the queueing system until that customers next arrival to the
queue, e.g., machine-repair problem, machines are customers
and a runtime is time to failure.
Let A
1
(i)
, A
2
(i)
, be the successive runtimes of customer i, and
S
1
(i)
, S
2
(i)
be the corresponding successive system times:

Queue Behavior and Queue Discipline

[Characteristics of Queueing System]
Queue behavior: the actions of customers while in a queue
waiting for service to begin, for example:
Balk: leave when they see that the line is too long,
Renege: leave after being in the line when its moving too slowly,
Jockey: move from one line to a shorter line.
Queue discipline: the logical ordering of customers in a queue
that determines which customer is chosen for service when a
server becomes free, for example:
First-in-first-out (FIFO)
Last-in-first-out (LIFO)
Service in random order (SIRO)
Shortest processing time first (SPT)
Service according to priority (PR).

Service Times and Service Mechanism

[Characteristics of Queueing System]
Service times of successive arrivals are denoted by S
1
,
S
2
, S
3
.
May be constant or random.
{S
1
, S
2
, S
3
, } is usually characterized as a sequence of
independent and identically distributed random variables, e.g.,
exponential, Weibull, gamma, lognormal, and truncated normal
distribution.
A queueing system consists of a number of service
centers and interconnected queues.
Each service center consists of some number of servers, c,
working in parallel, upon getting to the head of the line, a
customer takes the 1
st
available server.

Service Times and Service Mechanism

[Characteristics of Queueing System]
Example: consider a discount warehouse where
customers may:
Serve themselves before paying at the cashier:

Service Times and Service Mechanism

[Characteristics of Queueing System]
Wait for one of the three clerks:
Batch service (a server serving several customers
simultaneously), or customer requires several servers
simultaneously.

Queueing Notation
[Characteristics of Queueing System]
A notation system for parallel server queues: A/B/c/N/K
A represents the interarrival-time distribution,
B represents the service-time distribution,
c represents the number of parallel servers,
N represents the system capacity,
K represents the size of the calling population.

Queueing Notation
[Characteristics of Queueing System]
Primary performance measures of queueing systems:
P
n
: steady-state probability of having n customers in system,
P
n
(t): probability of n customers in system at time t,
: arrival rate,

e
: effective arrival rate,
: service rate of one server,
: server utilization,
A
n
: interarrival time between customers n-1 and n,
S
n
: service time of the nth arriving customer,
W
n
: total time spent in system by the nth arriving customer,
W
n
Q
: total time spent in the waiting line by customer n,
L(t): the number of customers in system at time t,
L
Q
(t): the number of customers in queue at time t,
L: long-run time-average number of customers in system,
L
Q
: long-run time-average number of customers in queue,
w: long-run average time spent in system per customer,
w
Q
: long-run average time spent in queue per customer.

Time-Average Number in System L

[Characteristics of Queueing System]
Consider a queueing system over a period of time T,
Let T
i
denote the total time during [0,T] in which the system
contained exactly i customers, the time-weighted-average number
in a system is defined by:
Consider the total area under the function is L(t), then,
The long-run time-average # in system, with probability 1:


=

=
|
.
|

\
|
= =
0 0
1

i
i
i
i
T
T
i iT
T
L

= =

=
T
i
i
dt t L
T
iT
T
L
0
0
) (
1 1

=

T L dt t L
T
L
T
as ) (
1

Time-Average Number in System L

[Characteristics of Queueing System]
The time-weighted-average number in queue is:
G/G/1/N/K example: consider the results from the queueing
system (N > 4, K > 3).

=
=
1 if , 1 ) (
0 if , 0
) (
L(t) t L
L(t)
t L
Q
customers 3 . 0
20
) 1 ( 2 ) 4 ( 1 ) 15 ( 0

=
+ +
=
Q
L
= =

=
T L dt t L
T
iT
T
L
Q
T
Q
i
Q
i Q
as ) (
1 1

0
0
cusomters 15 . 1 20 / 23
20 / )] 1 ( 3 ) 4 ( 2 ) 12 ( 1 ) 3 ( 0 [

= =
+ + + = L

Average Time Spent in System Per

Customer w [Characteristics of Queueing System]
The average time spent in system per customer, called
the average system time, is:
where W
1
, W
2
, , W
N
are the individual times that each of the N
customers spend in the system during [0,T].
For stable systems:
If the system under consideration is the queue alone:
G/G/1/N/K example (cont.): the average system time is
N w w as

=
=
N
i
i
W
N
w
1
1

1
1

as
N
Q
Q i Q
i
w W w N
N
=
=

units time 6 . 4
5
) 16 20 ( ... ) 3 8 ( 2
5
...

5 2 1
=
+ + +
=
+ + +
=
W W W
w

The Conservation Equation

[Characteristics of Queueing System]
Conservation equation (a.k.a. Littles law)
Holds for almost all queueing systems or subsystems (regardless
of the number of servers, the queue discipline, or other special
circumstances).
G/G/1/N/K example (cont.): On average, one arrival every 4 time
units and each arrival spends 4.6 time units in the system.
Hence, at an arbitrary point in time, there is (1/4)(4.6) = 1.15
customers present on average.
w L

=
= N T w L and as
Arrival rate
Average
System time
Average # in
system

Server Utilization
[Characteristics of Queueing System]
Definition: the proportion of time that a server is busy.
Observed server utilization, , is defined over a specified time
interval [0,T].
Long-run server utilization is .
For systems with long-run stability: T as

Server Utilization
[Characteristics of Queueing System]
For G/G/1// queues:
Any single-server queueing system with average arrival
rate customers per time unit, where average service time
E(S) = 1/ time units, infinite queue capacity and calling
population.
Conservation equation, L = w, can be applied.
For a stable system, the average arrival rate to the server,

s
, must be identical to .
The average number of customers in the server is:
( )
T
T T
dt t L t L
T
L
T
Q s
0
0
) ( ) (
1

= =

Server Utilization
[Characteristics of Queueing System]
In general, for a single-server queue:
For a single-server stable queue:
For an unstable queue ( > ), long-run server utilization is 1.

= =
= =
) ( and
as

s E
T L L
s s
1 < =

Server Utilization
[Characteristics of Queueing System]
For G/G/c// queues:
A system with c identical servers in parallel.
If an arriving customer finds more than one server idle, the
customer chooses a server without favoring any particular
server.
For systems in statistical equilibrium, the average number of
busy servers, L
s
, is: L
s
, = E(s) = /.
The long-run average server utilization is:
systems stable for where ,

c
c c
L
s
< = =

Server Utilization and System Performance

[Characteristics of Queueing System]
System performance varies widely for a given utilization .
For example, a D/D/1 queue where E(A) = 1/ and E(S) =
1/, where:
L = = /, w = E(S) = 1/, L
Q
= W
Q
= 0.
By varying and , server utilization can assume any value
between 0 and 1.
Yet there is never any line.
In general, variability of interarrival and service times
causes lines to fluctuate in length.

Server Utilization and System Performance

[Characteristics of Queueing System]
Example: A physician who schedules patients every 10 minutes and
spends S
i
minutes with the i
th
patient:
Arrivals are deterministic, A
1
= A
2
= =
-1
= 10.
Services are stochastic, E(S
i
) = 9.3 min and V(S
0
) = 0.81 min
2
.
On average, the physician's utilization = = / = 0.93 < 1.
Consider the system is simulated with service times: S
1
= 9, S
2
=
12, S
3
= 9, S
4
= 9, S
5
= 9, . The system becomes:
The occurrence of a relatively long service time (S
2
= 12) causes a
waiting line to form temporarily.

=
1 . 0 y probabilit with minutes 12
9 . 0 y probabilit with minutes 9
i
S

Costs in Queueing Problems

[Characteristics of Queueing System]
Costs can be associated with various aspects of the
waiting line or servers:
System incurs a cost for each customer in the queue, say at a rate
of $10 per hour per customer.
The average cost per customer is:
If customers per hour arrive (on average), the average cost
per hour is:
Server may also impose costs on the system, if a group of c
parallel servers (1 c ) have utilization r, each server imposes
a cost of $5 per hour while busy.
The total server cost is: $5*c.
Q
N
j
Q
j
w
N
W
* 10 $
* 10 $
1
=

=
W
j
Q
is the time
customer j spends
in queue

hour /

* 10 $

* 10 $
customer
* 10 $
hour
customer

Q Q
Q
L w
w
= =
|
|
.
|

\
|
|
.
|

\
|

Steady-State Behavior of Infinite-Population

Markovian Models
Markovian models: exponential-distribution arrival process
(mean arrival rate = ).
Service times may be exponentially distributed as well (M) or
arbitrary (G).
A queueing system is in statistical equilibrium if the probability
that the system is in a given state is not time dependent:
P( L(t) = n ) = P
n
(t) = P
n
.
Mathematical models in this chapter can be used to obtain
approximate results even when the model assumptions do not
strictly hold (as a rough guide).
Simulation can be used for more refined analysis (more faithful
representation for complex systems).

Steady-State Behavior of Infinite-Population

Markovian Models
For the simple model studied in this chapter, the steady-state
parameter, L, the time-average number of customers in the
system is:
Apply Littles equation to the whole system and to the queue alone:
G/G/c// example: to have a statistical equilibrium, a
necessary and sufficient condition is /(c) < 1.

=
=
0 n
n
nP L
Q Q
Q
w L
w w
L
w

=
= =
1
,

M/G/1 Queues [Steady-State of Markovian Model]

Single-server queues with Poisson arrivals & unlimited capacity.
Suppose service times have mean 1/ and variance
2
and = / <
1, the steady-state parameters of M/G/1 queue:
) 1 ( 2
) / 1 (
,
) 1 ( 2
) / 1 ( 1
) 1 ( 2
) 1 (
,
) 1 ( 2
) 1 (
1 , /
2 2 2 2
2 2 2 2 2 2
0

+
=

+
+ =

+
=

+
+ =
= =
Q
Q
w w
L L
P

M/G/1 Queues [Steady-State of Markovian Model]

No simple expression for the steady-state probabilities P
0
, P
1
,
L L
Q
= is the time-average number of customers being
served.
Average length of queue, L
Q
, can be rewritten as:
If and are held constant, L
Q
depends on the variability,
2
, of the
service times.
) 1 ( 2 ) 1 ( 2
2 2 2

=
Q
L

M/G/1 Queues [Steady-State of Markovian Model]

Example: Two workers competing for a job, Able claims to be faster
than Baker on average, but Baker claims to be more consistent,
Poisson arrivals at rate = 2 per hour (1/30 per minute).
Able: 1/ = 24 minutes and
2
= 20
2
= 400 minutes
2
:
The proportion of arrivals who find Able idle and thus experience no delay is P
0
= 1- = 1/5 = 20%.
Baker: 1/ = 25 minutes and
2
= 2
2
= 4 minutes
2
:
The proportion of arrivals who find Baker idle and thus experience no delay is
P
0
= 1- = 1/6 = 16.7%.
Although working faster on average, Ables greater service variability
results in an average queue length about 30% greater than Bakers.
customers 711 . 2
) 5 / 4 1 ( 2
] 400 24 [ ) 30 / 1 (
2 2
=

+
=
Q
L
customers 097 . 2
) 6 / 5 1 ( 2
] 4 25 [ ) 30 / 1 (
2 2
=

+
=
Q
L

M/M/1 Queues [Steady-State of Markovian Model]

Suppose the service times in an M/G/1 queue are
exponentially distributed with mean 1/, then the variance
is
2
= 1/
2
.
M/M/1 queue is a useful approximate model when service
times have standard deviation approximately equal to their
means.
The steady-state parameters:
( )
( )
( ) ) 1 (
,
) 1 (
1 1
1
,
1
1 , /
2 2

=
= =
Q
Q
n
n
w w
L L
P

M/M/1 Queues [Steady-State of Markovian Model]

Example: M/M/1 queue with service rate =10 customers
per hour.
Consider how L and w increase as arrival rate, , increases from 5
to 8.64 by increments of 20%:
If / 1, waiting lines tend to continually grow in length.
Increase in average system time (w) and average number in
system (L) is highly nonlinear as a function of .

5.0 6.0 7.2 8.64 10.0

0.500 0.600 0.720 0.864 1.000

L 1.00 1.50 2.57 6.35
w 0.20 0.25 0.36 0.73

Effect of Utilization and Service Variability

[Steady-State of Markovian Model]
For almost all queues, if lines are too long, they can be reduced
by decreasing server utilization () or by decreasing the service
time variability (
2
).
A measure of the variability of a distribution, coefficient of
variation (cv):
The larger cv is, the more variable is the distribution relative to its
expected value
| |
2
2
) (
) (
) (
X E
X V
cv =

Effect of Utilization and Service Variability

[Steady-State of Markovian Model]
Consider L
Q
for any M/G/1
queue:
|
|
.
|

\
|
+
|
|
.
|

\
|

+
=
2
) ( 1
1
) 1 ( 2
) 1 (

2 2
2 2 2
cv
L
Q

L
Q
for M/M/1
queue
Corrects the M/M/1
formula to account
for a non-exponential
service time distn

Multiserver Queue [Steady-State of Markovian Model]

M/M/c// queue: c channels operating in parallel.
Each channel has an independent and identical exponential
service-time distribution, with mean 1/.
To achieve statistical equilibrium, the offered load (/) must
satisfy / < c, where /(c) = is the server utilization.
Some of the steady-state probabilities:
( )

L
w
c L P
c
c c
P c
c L
c
c
c n
P
c
c
c
c
n
n
=

+ =

+ =

(
(

|
|
.
|

\
|

|
.
|

\
|
|
|
.
|

\
|
+
(
(

=
=
+

1
) (
) 1 )( ! (
) (
!
1
!
) / (
/
2
0
1
1
1
0
0

Multiserver Queue [Steady-State of Markovian Model]

Other common multiserver queueing models:
M/G/c/: general service times and c parallel server. The
parameters can be approximated from those of the M/M/c//
model.
M/G/: general service times and infinite number of servers, e.g.,
customer is its own system, service capacity far exceeds service
demand.
M/M/C/N/: service times are exponentially distributed at rate m
and c servers where the total system capacity is N c customer
(when an arrival occurs and the system is full, that arrival is turned
away).

Steady-State Behavior of Finite-Population

Models
When the calling population is small, the presence of one or
more customers in the system has a strong effect on the
distribution of future arrivals.
Consider a finite-calling population model with K customers
(M/M/c/K/K):
The time between the end of one service visit and the next call for
service is exponentially distributed, (mean = 1/).
Service times are also exponentially distributed.
c parallel servers and system capacity is K.

Steady-State Behavior of Finite-Population

Models
Some of the steady-state probabilities:

c L w nP L
K c c n
c c n K
K
c n P
n
K
P
c c n K
K
n
K
P
e e
K
n
n
n
c n
n
n
K
c n
n
c n
c
n
n
/ , / ,
,... 1 , ,
! )! (
!
1 ,..., 1 , 0 ,
! )! (
!
0
0
1
1
0
0
= = =

+ =
|
|
.
|

\
|

=
|
|
.
|

\
|
|
|
.
|

\
|
=

|
|
.
|

\
|

+
|
|
.
|

\
|
|
|
.
|

\
|
=

=
=
K
n
n e
e
P n K
0
) (
service) xiting entering/e (or queue to customers of rate arrival effective run long the is where

B17Slide 38
B17 typo
Brian; 2005/01/09

Steady-State Behavior of Finite-Population

Models
Example: two workers who are responsible for10 milling
machines.
Machines run on the average for 20 minutes, then require an
average 5-minute service period, both times exponentially
distributed: = 1/20 and = 1/5.
All of the performance measures depend on P
0
:
Then, we can obtain the other P
n
.
Expected number of machines in system:
The average number of running machines:
065 . 0
20
5
2 ! 2 )! 10 (
! 10
20
5 10
1
10
2
2
1 2
0
0
=

|
.
|

\
|

+ |
.
|

\
|
|
|
.
|

\
|
=

=

=

n
n
n
n
n
n
n
P
machines 17 . 3
10
0
= =

= n
n
nP L
machines 83 . 6 17 . 3 10 = = L K

Networks of Queues
Many systems are naturally modeled as networks of single
queues: customers departing from one queue may be routed
to another.
The following results assume a stable system with infinite
calling population and no limit on system capacity:
Provided that no customers are created or destroyed in the
queue, then the departure rate out of a queue is the same as the
arrival rate into the queue (over the long run).
If customers arrive to queue i at rate
i
, and a fraction 0 p
ij
1 of
them are routed to queue j upon departure, then the arrival rate
form queue i to queue j is
i
p
ij
(over the long run).

Networks of Queues
The overall arrival rate into queue j:
If queue j has c
j
< parallel servers, each working at rate
j
, then
the long-run utilization of each server is
j
=
j
/(c
j
) (where
j
< 1
for stable queue).
If arrivals from outside the network form a Poisson process with
rate a
j
for each queue j, and if there are c
j
identical servers
delivering exponentially distributed service times with mean 1/
j
,
then, in steady state, queue j behaves likes an M/M/c
j
queue with
arrival rate

+ =
i
ij i j j
p a
all

Arrival rate
from outside
the network
Sum of arrival rates
from other queues
in network

+ =
i
ij i j j
p a
all

Network of Queues
Discount store example:
Suppose customers arrive at the rate 80 per hour and 40%
choose self-service. Hence:
Arrival rate to service center 1 is
1
= 80(0.4) = 32 per hour
Arrival rate to service center 2 is
2
= 80(0.6) = 48 per hour.
c
2
= 3 clerks and
2
= 20 customers per hour.
The long-run utilization of the clerks is:

2
= 48/(3*20) = 0.8
All customers must see the cashier at service center 3, the
overall rate to service center 3 is
3
=
1
+
2
= 80 per hour.
If
3
= 90 per hour, then the utilization of the cashier is:

3
= 80/90 = 0.89

Summary
Introduced basic concepts of queueing models.
Show how simulation, and some times mathematical analysis, can
be used to estimate the performance measures of a system.
Commonly used performance measures: L, L
Q
, w, w
Q
, , and
e
.
When simulating any system that evolves over time, analyst must
decide whether to study transient behavior or steady-state behavior.
Simple formulas exist for the steady-state behavior of some queues.
Simple models can be solved mathematically, and can be useful in
providing a rough estimate of a performance measure.