Lecture Day 13

Queuing Models

Queuing Models

Queue is another term for a waiting line, and

a queuing system is simply a system that
involves a waiting line. Queuing theory is a
branch of management science that enables
the analyst to describe the behavior of
queuing systems.
Queuing theory does not address
optimization problems directly. Rather, it
uses elements of statistics and mathematics
for the construction of models that describe
the important descriptive statistics of a
queuing system.

Members of a queue are known as calling

units.
The queuing system descriptive statistics
include such factors as the expected waiting
time of the calling units, the expected length
of the line, and the percentage of idle time
for the service facility (the source of goods
or services for which the calling units wait).
When queuing theory is applied,
managements objective is usually to
minimize two kinds of costs:

Those associated with providing service

Those associated with waiting time

Parts of Any Queuing

System

Calling
Population

...
Queue

Service
Facility

Served
Calling
Units

Calling
Population

...

The calling population has three

characteristics that are important to
consider when deciding on what type of
queuing model to apply:
The size of the calling population

The pattern of arrivals at the queuing

system

The attitude of the calling units

Calling
Population

...

The calling population has three

characteristics that are important to
consider when deciding on what type of
queuing model to apply:
The size of the calling population can be
finite or infinite.
The key to determining whether an
infinite calling population can be
assumed is whether the probability of
an arrival is significantly changed when
a member or members of a population
are receiving service and thus cannot
arrive to the system.

Calling
Population

...

The calling population has three

characteristics that are important to
consider when deciding on what type of
queuing model to apply:
The size of the calling population

The pattern of arrivals at the queuing

system
can
be pre-determined/scheduled or
random.
If arrivals are scheduled, analytical
queuing models are usually
inappropriate. If arrivals are random, it
is necessary to determine the
probability distribution of the time
between intervals.

It has been shown mathematically that if the

probability density function of the interarrival times is exponential, calling units
arrive according to a so-called Poisson
process.
Poisson arrivals generally exist in situations
where the number of arrivals during a certain
time interval is independent of the number of
arrivals that have occurred in previous time
arrivals.

This basic property states that the conditional

probability of any future event depends only
on the present state of the system and is
independent of previous states of the system.

The Poisson Probability Density Function

gives the probability of n arrivals in time
period t.

Pn(t) = e-t(t)n

n = 0, 1, 2, . . .

n!
where:
n = number of arrivals
t = size of the time interval
= mean arrival rate per unit of time

Calling
Population

...

The calling population has three

characteristics that are important to
consider when deciding on what type of
queuing model to apply:
The size of the calling population

The pattern of arrivals at the queuing

system

can be
patient
or impatient.
The attitude
of the calling units
There are two forms of impatient
attitudes, namely: balking and
reneging.

In Summary:

Size

Finite

Infinite

Calling
Population
Characteristics

Arrival
Pattern

Attitude

PreRandom Patient
Determin
ed
Poisson

Other Balking

Impatien
t
Renegin
g

...

Service
Facility

The service facility has three basic

properties:
The structure of the queuing system

The distribution of service times

The service discipline

...

Service
Facility

The service facility has three basic

properties:
The structure of the queuing system
can be single-phase or multi-phase.
The great majority of queuing models are
single-phase models. It is possible,
nonetheless, to view a multiphase system
as separate, single-phase systems in
which the output from one server
becomes the input for another server.

Possible
Structures:

Calling
Population

...

Service
Facility

Served
Calling
Units

Queue

Single-Phase, Single-Channel Queuing System

Possible
Structures:

Calling
Population

Service
Facility
no. 1
...

Queue

Service
Facility
no. 2

Served
Calling
Units

Single-Phase, Multi-Channel Queuing System

Possible
Structures:

Calling
Population

...

Queue

Service
Facility
type 1

Service
Facility
type 2

Served
Calling
Units

Queue

Multi-Phase, Single-Channel Queuing System

Possible
Structures:

Calling
Population

...

Queue

Servic
e
Facility

Servic
e
Facility

no.1
type 1
Servic
e
Facility

no.1
type 2
Servic
e
Facility

no.2
type 1

no.2
type 2

Served
Calling
Units

Multi-Phase, Multi-Channel Queuing System

...

Service
Facility

The service facility has three basic

properties:
The structure of the queuing system
can be
The distribution of service times
constant or random.
If service time is a random variable, it
is necessary to determine how that
random variable is distributed. In most
cases, service times are exponentially
distributed. As such, the probability of
relatively long service times is small.

...

Service
Facility

The service facility has three basic

properties:
The structure of the queuing system

The distribution of service times

determines which
The service discipline
calling unit in the queuing system
receives service.

Classifications of Service Disciplines

First come, first served

Priority

Preemptive

Non-preemptive

Random

In Summary:

Structure

Single
Phase

MultiPhase

Service
Facility
Characteristics

Service
Times

Service
Discipline

Constant Random FCFS PriorityRandom

Single
Multi- Exponenti
Channel Channel
al

Other Preemptiv
None
Preemptiv

Elementary
Queuing Models

Notations to be used:

Lq =

expected or mean length of the queue

(number of calling units in the

queue)

Ls =
expected number of calling units in
the
system (number in the
queue plus number
being served)

Wq=
in line

Ws=
expected or mean time spent in the
system
(including waiting time and
service time)

expected or mean time spent waiting

Notations to be used (cont.):

=
units

mean arrival rate (number of calling

per unit of time)

=
units

mean service rate (number of calling

served per unit of time)

1/ =

mean service time for a calling unit

s =
number of parallel (equivalent)
service
facilities in the system

P(n) = probability of having n units in the

system

server utilization factor (that is, the

proportion of time the server
can be
expected to be busy)

The Basic Single-Server Model

The assumptions of this model
are:
Poisson arrival process

Exponential service
times

Single server

FCFS service discipline

Infinite source

Infinite queue

Patient calling units

The Basic Single-Server Model

Statistics to be determined:
The probability of 0 calling units in the
system:
P(0) = 1 (/)

The probability of n calling units in the

system:
P(n) = P(0)(/)n

The proportion of time the server is busy:

= /

The Basic Single-Server Model

Statistics to be determined
(cont.):
Expected number of calling units in the
system:
L = / (-)
s

Expected number of calling units in the

queue:
Lq = 2/ [(-)]

Expected time in the system:

Ws = 1 / (-)

Expected time waiting in line:

Wq = / [(-)]

Example:
A particular toll road has one attendant
at an exit lane. Cars arrive at that toll gate in
a Poisson fashion at a rate of 120 cars per
hour, and it takes the attendant, on the
average, 15 seconds to service a car. Service
times are exponentially distributed.
Assumptions of an infinite calling population
and an infinite queue are reasonable. What
are the basic queuing statistics for this
system?

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.

The Basic Single-Server Model

Statistics to be determined:
The probability of 0 calling units in the
system:
P(0) = 1 (/)

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.
P(0) = 1 (/)
= 1 (/)
= 1 (/)
= /
=

The Basic Single-Server Model

Statistics to be determined:
The probability of 0 calling units in the
system:
P(0) = 1 (/)

The probability of n calling units in the

system:
P(n) = P(0)(/)n

The proportion of time the server is busy:

= /

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.
P(0) = 50%

= /
= 2/4
= 1/2
= 50%

The Basic Single-Server Model

Statistics to be determined
(cont.):
Expected number of calling units in the
system:
L = / (-)
s

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.
P(0) = 50%

= 50%
Ls = / (-)
= 2 / (4 2)
=2/2
= 1 car

The Basic Single-Server Model

Statistics to be determined
(cont.):
Expected number of calling units in the
system:
L = / (-)
s

Expected number of calling units in the

queue:
Lq = 2/ [(-)]

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.
P(0) = 50%

= 50%
Ls = 1 car
Lq = 2/ [(-)]
= 22 / [4 (4 2)]
= 4 / [4 (2)]
= 1/2 car

The Basic Single-Server Model

Statistics to be determined
(cont.):
Expected number of calling units in the
system:
L = / (-)
s

Expected number of calling units in the

queue:
Lq = 2/ [(-)]

Expected time in the system:

Ws = 1 / (-)

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.
P(0) = 50%

= 50%
Ls = 1 car
Lq = 1/2 car

Ws = 1 / (-)
= 1 / (4 2)
= 1/2 min.
= 30 seconds

The Basic Single-Server Model

Statistics to be determined
(cont.):
Expected number of calling units in the
system:
L = / (-)
s

Expected number of calling units in the

queue:
Lq = 2/ [(-)]

Expected time in the system:

Ws = 1 / (-)

Expected time waiting in line:

Wq = / [(-)]

 = 120 cars/hr.
= 15 sec./car
= 2 cars/min.
= 4 cars/min.
P(0) = 50%

= 50%
Ls = 1 car
Lq = 1/2 car

Ws = 1 / (-)
= 1 / (4 2)
= 1/2 min.
= 30 seconds

Wq = / [(-)]
= 2 / [4 (4 2)]
= 2 / [4 (2)]
= 1/4 min.
= 15 seconds

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
The assumptions of this model are
identical to those of the basic singleserver model except that the number of
servers is assumed to be greater than
one. Also, it is assumed that all servers
have the same rate of service.

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined:
The probability of 0 calling units in the
system:
P(0) =
1
s-1

n!

n=0

-1

(1- )

(/)n + (/)s
s!
s

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
The probability of n calling units in the
system:
P(n) = (/)n P(0) for 0 < n < s
n!
= (/)n P(0)
s!sn-s

for n > s

The proportion of time the server is busy:

= /s

assuming each server has the same

mean service rate of units per time
period

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
Expected number of calling units in the
queue:
Lq = P(0)(/)s
s!(1 )2

Expected number of calling units in the

system:
Ls = Lq + (/)

Expected time waiting in line:

Wq = Lq/

Expected time in the system:

Ws = Wq + (1/)

Example:
A particular toll road has three
attendants at an exit lane. Cars arrive at that
toll gate in a Poisson fashion at a rate of 600
cars per hour, and it takes the attendant, on
the average, 15 seconds to service a car.
Service times are exponentially distributed.
Assumptions of an infinite calling population
and an infinite queue are reasonable. What
are the basic queuing statistics for this
system?

 = 600 cars/hr.
= 15 sec./car
= 10 cars/min.
= 4 cars/min.
s = 3

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined:
The probability of 0 calling units in the
system:
P(0) =
1
s-1

n!

n=0

-1

(1- )

(/)n + (/)s
s!
s

 = 600 cars/hr. P(0) =

= 15 sec./car
= 10 cars/min.
= 4 cars/min.

n!

s = 3

1
s-1

-1

(1- )

(/)n + (/)s
s!
s

n=0

1
2

10 -1

(1- )

(10/4)n + (10/4)3
n! 3!

3(4)
n=0

P(0) =

1
2

10 -1

(1- )

(10/4)n + (10/4)3
n! 3!

3(4)
n=0
1
1
+
=

(10/4)
+

10 -1

(1- )

(10/4)2 + (10/4)3
12
2 6
1

(10/4) (10/4) + (10/4)3

+
2
=
1
=
= 4.50%
0.04494382
22.25
1
+

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
The probability of n calling units in the
system:
P(n) = (/)n P(0) for 0 < n < s
n!
= (/)n P(0)
s!sn-s

for n > s

The proportion of time the server is busy:

= /s

assuming each server has the same

mean service rate of units per time
period

 = 600 cars/hr.
= 15 sec./car
= 10 cars/min.
= 4 cars/min.
s = 3
P(0) =
0.04494382
= /s
= 10/[ 3(4) ]
= 10 / 12
= 5/6
= 0.83
= 83.33%

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
Expected number of calling units in the
queue:
Lq = P(0)(/)s
s!(1 )2

 = 600 cars/hr.
= 15 sec./car
= 10 cars/min.
= 4 cars/min.
s = 3
P(0) =
0.04494382
= 0.83

Lq = P(0)(/)s
s!(1 )2
= (0.04494382) (10/4)3 (0.83)
3! (1 0.83)2
= (0.04494382) (15.625)
(0.83)
(6) (1 0.83)2
= 0.58520599
0.16
= 3.51123595 car

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
Expected number of calling units in the
queue:
Lq = P(0)(/)s
s!(1 )2

Expected number of calling units in the

system:
Ls = Lq + (/)

 = 600 cars/hr.
= 15 sec./car
= 10 cars/min.
= 4 cars/min.

Ls = Lq + (/)
= 3.51123595 +
(10/4)
= 3.51123595 + 2.5
= 6.01123595 or 6 cars

s = 3
P(0) =
0.04494382
= 0.83
Lq = 3.51123595 cars

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
Expected number of calling units in the
queue:
Lq = P(0)(/)s
s!(1 )2

Expected number of calling units in the

system:
Ls = Lq + (/)

Expected time waiting in line:

Wq = Lq/

 = 600 cars/hr.
= 15 sec./car
= 10 cars/min.
= 4 cars/min.

Wq = Lq/
= 3.51123595 /
10
= 0.35112359
min.
= 21.07 sec.

s = 3
P(0) =
0.04494382
= 0.83
Lq = 3.51123595 cars
Ls = 6.01123595 cars

Multi-Server Model with Poisson

Arrivals and Exponential Service
Times
Statistics to be determined
(cont.):
Expected number of calling units in the
queue:
Lq = P(0)(/)s
s!(1 )2

Expected number of calling units in the

system:
Ls = Lq + (/)

Expected time waiting in line:

Wq = Lq/

Expected time in the system:

Ws = Wq + (1/)

 = 600 cars/hr.
= 15 sec./car
= 10 cars/min.
= 4 cars/min.

Wq = Lq/
= 3.51123595 /
10
= 0.35112359
min.
= 21.07 sec.

s = 3
Ws = Wq + (1/)
P(0) =
= 0.35112359 +
0.04494382
= 0.83
(1/4)
= 0.60112359 min.
Lq = 3.51123595 cars
= 36.07 sec.
Ls = 6.01123595 cars

End of Day 13