C h a p t e r 16

Queuing Theory
“Good managers have a bias for action.”
– Peters, Thomas J.

PREVIEW
A queue, in general, is formed at any place when a customer (human beings or physical entities) that
requires service is made to wait due to the fact that the number of customers exceeds the number of
service facilities or when service facilities do not work efficiently and take more time than prescribed to
serve a customer.

LEARNING OBJECTIVES
After studying this chapter, you should be able to
z identify and examine situations that generate queuing problems.
z describe the trade-off between cost of service and cost of waiting time.
z understand various components (or parts) of a queuing system and description of each of them.
z analyze the variety of performance measures (operating characteristics) of a queuing system.
z derive relationship among variety of performance measures using probability distributions.
z make distinction between several queuing models and derive performance measures for each of them.

CHAPTER OUTLINE
16.1 Introduction 16.8 Finite Calling Population Queuing Models
16.2 The Structure of a Queuing System • Self Practice Problems C
16.3 Performance Measures of a Queuing System 16.9 Multi-phase Service Queuing Model
16.4 Probability Distributions in Queuing Systems • Self Practice Problems D
• Conceptual Questions A • Hints and Answers
16.5 Classification of Queuing Models 16.10 Special Purpose Queuing Models
16.6 Single-server Queuing Models ‰ Chapter Summary
• Conceptual Questions B ‰ Chapter Concepts Quiz
• Self Practice Problems A ‰ Case Study
• Hints and Answers ‰ Appendix 16.A : Probability Distribution of Arrivals
16.7 Multi-server Queuing Models and Departures
• Conceptual Questions C ‰ Appendix 16.B : Erlangian Service Time Distribution
• Self Practice Problems B with K-Phases
• Hints and Answers
560 Operations Research: Theory and Applications

16.1 INTRODUCTION
A common situation that occurs in everyday life is that of waiting in a line either at bus stops, petrol pumps,
restaurants, ticket booths, doctors’ clinics, bank counters, traffic lights and so on. Queues (waiting lines)
are also found in workshops where the machines wait to be repaired; at a tool crib where the mechanics
wait to receive tools; in a warehouse where items wait to be used, incoming calls wait to mature in the
telephone exchange, trucks wait to be unloaded, airplanes wait either to take off or land and so on.
In general, a queue is formed at a production/operation system when either customers (human beings
or physical entities) requiring service wait because number of customers exceeds the number of service
facilities, or service facilities do not work efficiently/take more time than prescribed to serve a customer.
Queuing theory can be applied to a variety of situations where it is not possible to accurately predict
the arrival rate (or time) of customers and service rate (or time) of service facility or facilities. In particular,
it can be used to determine the level of service (either the service rate or the number of service facilities)
that balances the following two conflicting costs:
(i) cost of offering the service
(ii) cost incurred due to delay in offering service
The first cost is associated with the service facilities and their operation, and the second represents the
cost of customers waiting for service.
Obviously, an increase in the existing service facilities would reduce the customer’s waiting time.
The study of Conversely, decreasing the level of service would result in long queue(s). This means an increase in the
queuing theory level of service increases the cost of operating service facilities but decreases the cost of customers waiting
helps to determine
the balance between for service. Figure 16.1 illustrates both types of costs as a function of level of service. The combined cost
• cost of offering of service and customer waiting cost is U-shaped because of their trade-off relationship. The total cost is
the service, and minimized at the lowest point of the total cost curve. The service level is one that minimizes the sum of
• cost incurred due the two costs.
to delay in
offering service

Fig. 16.1
Operational Cost
vs Level of
Service

Since customer waiting cost for service is difficult to estimate, it is usually measured in terms of loss
of sales or goodwill when the customer is a human being and has no sympathy with the service system.
But, if the customer is a machine waiting for repair, then cost of waiting is measured in terms of cost of
lost production.
Many real-life situations in which study of queuing theory can provide solution to waiting line
problems are listed in Table 16.1.

Situation Customers Service Facilities

Petrol pumps (stations) Automobiles Pumps/Passionel
Hospital Patients Doctors/Nurses/Rooms
Airport Aircraft Runways
Post office Letters Sorting system
Job interviews Applicants Interviewers
Table 16.1 Cargo Trucks Loader/unloaders
Examples of Workshop Machines/Cars Mechanics/Floor space
Queues Factory Employees Cafeteria/Punching Machine
 Queuing Theory 561

16.2 THE STRUCTURE OF A QUEUING SYSTEM

The major components (parts or elements) of any waiting-line (queuing) system are shown in Fig. 16.2. Each
of these components is discussed below:
1. Calling population (or input source)
2. Queuing process
3. Queue discipline
4. Service process (or mechanism)

Fig. 16.2
Components of a
Queuing System

Potential customers who arrive to the queuing system is referred as calling population, also known as
customer (input) source.
The manner in which customers arrive at the service facility, individually, or in batches, at scheduled
or unscheduled time is called the arrival process. The customer’s entry into the queuing system depends
upon the queue conditions.
Customers, from a queue, are selected for service according to certain rules known as queue discipline.
A service facility may be without server (self service), or may include one or more servers operating either
in a series (as a team) or in parallel (multiple service channels). The rate (constant or random) at which service
is rendered is known as the service process. After the service is rendered, the customer leaves the system.
If the server is idle at the time of the customer’s arrival, then the customer is served immediately,
otherwise the customer is asked to join a queue or waiting line, which may have single, multiple or even
priority lines.

16.2.1 Calling Population Characteristics

The arrivals or inputs to the system are characterized by:
• Size of calling population
• behaviour of the arrivals
• pattern of arrivals at the system An infinite calling
The calling population need not be homogeneous and may consist of several subpopulations. For population is one
where a customer’s
example, patients arriving at the OPD of a hospital are usually of three categories: walk-in patients, patients arrival is
with appointments and emergency patients. Each patient class places different demands on service facility, independent of
but the waiting expectations of each category differ significantly. customers already
present in the
Size of calling population If probability of an arrival depends on the number of customers already system.
in the system (in service plus in queue), the calling population is called limited or finite. Examples of finite
population are queuing systems that have limited access for service, such as (i) machine operator
responsible only to handle 5 machines (say), (ii) salesman responsible to handle limited number of
customers, etc.
Alternately, if probability of an arrival does not depend on the number of customers already in the
system, the calling population is called unlimited or infinite. Examples of infinite population are those open
to the general public, such as banks, supermarkets, petrol pump, ticket counter, cinema halls, restaurants,
etc., because in such a queuing system customers already present do not decrease potential of others in
the population to enter the system.
Behaviour of arrivals If a customer, on arriving at the service system waits in the queue until served and
does not switch between waiting lines, he is called a patient customer. Machines arrived at the maintenance
shop are examples of patient customers. Whereas the customer, who waits for a certain time in the queue and
leaves the service system without getting service due to certain reasons is called an impatient customer. For
example, a customer who has just arrived at a grocery store and finds that the salesmen are busy in serving
the customers already in the system, will either wait for service till his patience is exhausted or estimates that
his waiting time may be excessive and so leaves immediately to seek service elsewhere.
562 Operations Research: Theory and Applications

• Balking Customers do not join the queue either by seeing the number of customers already in
service system or by estimating the excessive waiting time for the desired service.
• Reneging Customers, after joining the queue, wait for sometime in the queue but leave before
being served on account of certain reasons.
• Jockeying Customers move from one queue to another hoping to receive service more quickly
(a common scene at a railway booking window).
Pattern of arrivals at the system Customers may arrive in batches (such as the arrival of a family at
a restaurant) or individually (such as the arrival of a train at a platform). These customers may arrive at a
service facility either on scheduled time (by prior information) or on unscheduled time (without information).
The arrival process (or pattern) of customers to the service system is classified into two categories:
static and dynamic. These two are further classified based on the nature of arrival rate and the control that
can be exercised on the arrivals.
The static arrival process in controlled by the nature of arrival rate (random or constant): In random
(or unscheduled) arrivals the time of arrival is a random variable and therefore need to understand the
average and frequency distribution of the times. In both the cases, the arrival process can be described
either by the average arrival rate (average number of arrivals per unit of time) or by the average
interarrival time (average time between two consecutive arrivals).
The dynamic arrival process is controlled by both the service facility and the customers. The service
facility adjusts its capacity by either varying manpower at different timings of service, varying service
charges (telephone call charges at different hours of the day or week) at different timings, or allowing entry
with appointments to match changes required in the service intensity.
The variation in the service intensity affects the customer’s behaviour. Customers either balk or renege
from the service system on seeing a long or slow moving waiting line.
The arrival time distribution can be approximated by one of the following probability distributions:
• Poisson distribution
• Exponential distribution
• Erlang distribution
The behaviour of
the arrivals at any The Poisson distribution, a discrete probability distribution, describes the arrival rate variability, i.e., number
queuing system is of random arrivals at a service facility in a fixed period of time. Another probability distribution that
categoried as: describes the average time between arrivals (inter-arrival time) when arrival rate is Poisson is called
• Balking exponential probability distribution. Let n customers arrive in a time interval 0 to t. If λ is the expected
• Reneging
(or average) number of arrivals per time unit, then the expected number of arrivals in a given time interval
• Jockeying
0 to t will be λt. Then the Poisson probability distribution function is given by:

(λt )n e −λ t
P (x = n) = , for n = 0, 1, 2, . . .
n!
This equation, can be used to determine the probability of no arrival in the given time interval 0 to t:

(λ t ) 0 e − λ t
P( x = 0) = = e −λ t
0!
Let random variable T represent the time between successive arrivals. Since a customer can arrive at
any time, therefore T must be a continuous random variable. The probability of no arrival in the time interval
0 to t will be equal to the probability that T exceeds t. Thus
P(T > t ) = P ( x = 0) = e − λt
The cumulative probability that the time T between two successive arrivals is t or less is given by:

P(T ≤ t ) = 1 − P(T > t ) = 1− e− λ t ; t ≥0

The expression for P(T ≤ t) is also called the cumulative probability distribution function of T. Also the
distribution of the random variable T is referred to as the exponential distribution, whose probability
density function can be written as follows:
 Queuing Theory 563

λ e −λ t , for λ , t ≥ 0
f (t ) = 
0 , otherwise
A typical exponential probability distribution function and cumulative exponential probability distribution
function is illustrated in Figs. 16.3(a) and (b), respectively.
As discussed earlier, if the Poisson distribution describes arrival of customers at a service system, then
the exponential distribution describes the time between successive arrivals. Thus for a Poisson distribution,
the average (or mean), µ and standard deviation, σ are equal, i.e. µ = σ = λ. Therefore, for an exponential
distribution, the average (mean), µ and standard deviation, σ are µ = σ = 1/λ.
Illustration Let on an average 4 customers arrive after every 2 minutes. Then number of arrivals per minute
will be, λ = 4/2 = 2.
(i) Probability of no more than 2 minutes gap between successive arrivals is:
P(T ≤ 2) = 1 – e–2(2) = 1 – e–4 = 1 – 0.0183 = 0.9817
(ii) Average time between successive arrivals,
1 1
= = 0.5 (= 30 seconds)
λ 2
(iii) Probability of interarrival time between successive arrivals is:
P(T ≤ 0.5) = 1– e–2(0.5) = 1 – e–1
= 1 – 0.606 = 0.394
Also, if λ = 24 customers arrive on an average rate of every 60 minutes or λ = 24/60 = 0.4 arrivals/min, and
a customer has already arrived, then the probability of one customer arriving in the next 5 minutes is given
by:
P(T < 5) = 1 – e – λt = 1 – e – 0.4(5) = 1 – e – 2 = 1 – 0.1353 = 0.8647
Further, if λ = 24 customers arrive on an average of every one hour, then interarrival time, t = 5 minutes
= 1/12 hour and λ t = 24×(1/12) = 2, then probability that there are 2 customers in the system is given by:

e −λt (λt ) 2 e−2 (2)2

P(x = 2) = = = 2e–2 = 0.2706
2! 2!

(a) Exponential Probability Distribution (b) Cumulative Exponential Fig. 16.3

Probability Distribution

16.2.2 Queuing Process

The queuing process refers to the number of queues – single, multiple or priority queues and their lengths.
The type of queue depends on the layout of service mechanism and the length (or size) of a queue depends
upon operational situations such as physical space, legal restrictions, and attitude of the customers.
In certain cases, a service system is unable to accommodate more than the required number of
customers at a time. No further customers are allowed to enter until more space is made available to
accommodate new customers. Such type of situations are referred to as finite (or limited) source queue.
Examples of finite source queues are cinema halls, restaurants, etc. On the other hand, if a service system
564 Operations Research: Theory and Applications

is able to accommodate any number of customers at a time, then it is referred to as infinite (or unlimited)
source queue. For example, in a sales department where the customer orders are received, there is no
restriction on the number of orders that can come in.
On arriving at a service system, if customers find long queue(s) in front of a service facility, then they
The queuing often do not enter the service system inspite additional waiting space is available. The queue length in such
process refers to cases depends upon the attitude of the customers. For example, when a motorist finds that there are many
the number of
vehicles waiting at the petrol station, in most of the cases, he does not stop at this station and seeks service
queues and their
respective lengths. elsewhere.
In some finite source queuing systems, no queue is allowed to form. For example, when a parking space
(service facility) cannot accommodate additional incoming vehicles (customers), the motorists are diverted
elsewhere.
Multiple queues at a service facility can also be finite or infinite. But this has certain advantages such as:
• Division of manpower is possible.
• Customer has the option of joining any queue.
• Balking behaviour of the customers can be controlled.

16.2.3 Queue Discipline

The queue discipline is the order (or manner) in which customers from the queue are selected for service.
There are a number of ways in which customers in the queue are served. Some of these are:
(a) Static Queue Disciplines These are based on the individual customer’s status in the queue. Few
of such queue disciplines are:
(i) If customers are served in the order of their arrival, then this is known as the first-come, first-
served (FCFS) service discipline. Prepaid taxi queue at airports where a taxi is engaged on a
‘first-come, first-served’ basis is an example of this discipline.
(ii) Other discipline which is also common in use is last-come, first-served (LCFS). This discipline
is mostly practised in (i) cargo handling where the last item loaded is removed first because
it reduces handling and transportation cost, and (ii) production process where items arrive at
a workplace are stacked one on top of the other and item that is the last one to have arrived
for service is processed first.
(b) Dynamic Queue Disciplines These are based on the individual customer attributes in the queue.
Few of such queue disciplines are:
(i) Service in random Order (SIRO) : Under this rule customers are selected for service at
random, irrespective of their arrivals in the service system.
(ii) Priority service : Under this rule customers are grouped in priority classes on the basis of
some attributes such as service time or urgency. The FCFS rule is used within each class to
provide service. The payment of telephone or electricity bills by cheque or cash are examples
of this discipline.
(iii) Pre-emptive priority (or Emergency) : Under this rule, an important customer is allowed to
Queue discipline enter into the service immediately after entering into the system, even if a customer with lower
refers to the priority is already in service. That is, lower priority customer’s service is interrupted (pre-
selection of
customers from a
empted) to start the service for such a customer. This interrupted service is resumed after the
queue for service. priority customer is served.
(iv) Non-pre-emptive priority : In this case an important customer is allowed to go ahead in the
queue, but the service his started immediately on completion of the current service.

16.2.4 Service Process (or Mechanism)

The service mechanism (or process) is concerned with the manner in which customers are serviced and
leave the service system. It is characterized by:
z The arrangement (or capacity) of service facilities

z The distribution of service times

z Server’s behaviour

z Management policies
 Queuing Theory 565

Arrangement of Service Facilities

The capacity of the service facility is measured in terms of customers who can be served simultaneously
and/or effectively. The service facilities (or servers), commonly known as service channels, may be in series,
in parallel or mixed (partly in series and parallel).

Fig. 16.4
Arrangements of
Service Facilities
in Series

The series arrangement consists of number of service facilities in the sequence so that a customer
goes through each facility, in a particular sequence, before the service is completed. Each service facility
may, however, work independently having its own rules of service. For example, during university/college
admissions, the students go through one counter after another for completing their admission formalities.
Figure 16.4 illustrates the arrangement of service facilities in series.
The parallel arrangement consists of a number of service facilities in parallel so that a customer may
join the queue of choice in front of service facilities or may be served by any service facility. The number
of check-in counters at an airport or ticket counters at railway stations are few examples of the arrangement
of service facilities in parallel. These are parallel because the customer may check-in or get ticket from any
counter. Figures 16.5(a) and (b) illustrate the arrangement of service facilities in parallel.

Fig. 16.5
Arrangement of
Service Facilities
(a) Single Queue, Multiple Service Facilities in Parallel (b) Multiple Queues, Multiple Service Facilities in in Parallel
Parallel

The mixed arrangement consists of service facilities arranged in series as well as in parallel as shown
in Fig. 16.6. For example, in a hospital, an incoming patient may need to go to any of the OPD counters,
where he may be checked (served) by various doctors (facilities in series) one by one.
Fig. 16.6
Single Queue,
Multiple Service
Facilities in
Parallel and in
Series

Service Time Distribution

The time taken by the server from the commencement of service to the completion of service for a customer
is known as the service time. A random service time may be described in two ways:
(a) Average Service Rate : The service rate measures the service capacity of the facility in terms of
customers per unit of time. If µ is the average service rate, then the expected number of customers
566 Operations Research: Theory and Applications

served during time interval 0 to t will be µt. If the service time is exponentially distributed, then the
service rate is described by Poisson distributed. If service starts at zero time, the probability that service
is not completed by time t is given by:
P(x = 0) = e – µt
If the random variable T represents the service time, then the probability of service completion within
time t is given by:
P(T ≤ t) = 1 – e – µt, t ≥ 0
(b) Average Length of Service Time : The fluctuating service time is described by the negative exponential
probability distribution, and is denoted by 1/µ.

16.3 PERFORMANCE MEASURES OF A QUEUING SYSTEM

The performance measures (operating characteristics) for the evaluation of the performance of an existing
queuing system, and for designing a new system in terms of the level of service a customer receives as
well as the proper utilization of the service facilities are listed as follows:
Service time is the 1. Average (or expected) time spent by a customer in the queue and system
elapsed time from
the beginning to the Wq : Average time an arriving customer has to wait in a queue before being served,
end of a customer’s Ws : Average time an arriving customer spends in the system, including waiting and service.
service.
2. Average (expected) number of customers in the queue and system
Lq : Average number of customers waiting for service in the queue (queue length)
Ls : Average number of customers in the system (either waiting for services in the queue or being
served).

3. Value of time both for customers and servers

Pw : Probability that an arriving customer has to wait before being served (also called blocking
probability).
λ
ρ= : Percentage of time a server is busy serving customers, i.e., the system utilization.
µ
Pn : Probability of n customers waiting for service in the queuing system.
Pd : Probability that an arriving customer is not allowed to enter in the queuing i.e., system
capacity is full.

Steady-state
4. Average cost required to operate the queuing system
condition is the • Average cost required to operate the system per unit of time?
normal condition that
a queuing system is
• Number of servers (service centres) required to achieve cost effectiveness?
in after operating for
some time with a 16.3.1 Transient-State and Steady-State
fixed utilization
factor less than At the beginning of service operations, a queuing system is influenced by the initial conditions, such as
one. number of customers waiting for service and percentage of time servers are busy serving customers, etc.
This initial period is termed as transient-state. However, after certain period of time, the system becomes
independent of the initial conditions and enters into a steady-state condition.
To quantify various measures of system performance in each queuing model, it is assumed that the
system has entered into a steady-state.
Let Pn (t) be the probability that there are n customers in the system at a particular time t. Any change
in the value of Pn (t), with respect to time t, is denoted by Pn′ (t). In the case of steady-state, we have:
lim Pn (t ) = Pn (independent of time, t)
t→ ∞

or lim
t→∞
d
dt
l q
Pn (t ) =
d
dt
( Pn )

or lim Pn′( t ) = 0
t→∞
 Queuing Theory 567

If the arrival rate of customers at the system is more than the service rate, then a steady-state cannot
be reached, regardless of the length of the elapsed time.
Queue size, also referred as line length represents average number of customers waiting in the system
for service.
Queue length represents average number of customers waiting in the system and being served.
Notations The notations used for analyzing of a queuing system are as follows:
n= number of customers in the system (waiting and in service)
Pn = probability of n customers in the system
λ= average customer arrival rate or average number of arrivals per unit of time
in the queuing system
µ = average service rate or average number of customers served per unit time at the place of
service
λ Average service completion time ( 1/ µ )
= ρ= Average interarrival time ( 1/ λ )
µ
ρ = traffic intensity or server utilization factor Utilization factor is
P0 = probability of no customer in the system the average fraction
s = number of service channels (service facilities or servers) of time that the
servers are being
N = maximum number of customers allowed in the system utilized while serving
Ls = average number of customers in the system (waiting and in service) customers.
Lq = average number of customers in the queue (queue length)
Ws = average waiting time in the system (waiting and in service)
Wq = average waiting time in the queue
Pw = probability that an arriving customer has to wait (system being busy), 1 – P0 = (λ/µ)
For achieving a steady-state condition, it is necessary that, λ /µ < 1 (i.e. the arrival rate must be less
than the service rate). Such a situation arises when the queue length is limited, generally because of space,
capacity limitation or customers balk.

16.3.2 Relationships among Performance Measures

The following basic relationships hold for all infinite source queuing models.
∞ ∞
Ls = Σ n Pn and Lq = Σ ( n − s ) Pn
n=0 n=s
The general relationship among various performance measures is as follows:
(i) Average number of customers in the system is equal to the average number of customers in queue
(line) plus average number of customers being served per unit of time (system utilization).
Ls = Lq + Customer being served
λ
= Lq +
µ
λ
The value of ρ = is true for a single server finite-source queuing model.
µ
(ii) Average waiting time for a customer in the queue (line)
Lq
Wq =
λ
(iii) Average waiting time for a customer in the system including average service time
1
Ws = Wq +
µ
(iv) Probability of being in the system (waiting and being served) longer than time t is given by:
P(T > t) = e–(µ – λ) λ t and P(T ≤ t) = 1 – P(T > t)
where T = time spent in the system
t = specified time period
e = 2.718
568 Operations Research: Theory and Applications

(v) Probability of only waiting for service longer than time t is given by:
λ – (µ – λ)t
P(T > t) = e
µ
(vi) Probability of exactly n customers in the system is given by:
n n
λ  λ  λ 
Pn = P0   = 1 −   
µ  µ  µ 
(vii) Probability that the number of customers in the system, n exceeds a given number, r is given by:
r +1
λ
P(n > r) =  
µ
The general relationships among various performance measures are:
1 1
(a) Ls = λ Ws (b) Ws = Wq + = Ls
µ λ
λ 1 1
(c) Lq = Ls – = λ Wq (d) Wq = Ws − = Lq
µ µ λ
∞ Ls 1
(e) Ls = Σ nPn → Ws = → Wq = Ws − → Lq = λWq
n=0 λ µ

16.4 PROBABILITY DISTRIBUTIONS IN QUEUING SYSTEMS

It is assumed that customers arrive in random order at queuing system and their arrival can be described
by the Poisson distribution and inter-arrival times are described by exponential distribution.
In most cases, service time for customers is assumed to be exponentially distributed. This implies that the
probability of service completion is constant and independent of the length of service time.
The number of arrivals and departures (those served) during an interval of time in a queuing system
is controlled by the following assumptions (also called axioms):
(i) The probability of an event (arrival or departure) occurring during the time interval (t, t + ∆t)
depends upon the length of time interval ∆t. That is, the probability of the fact that the event does
not depend either on number of events that occur up to time t or the specific value of t (meaning
that the events that occur in non-overlapping time) are statistically independent.
(ii) The probability of more than one event occurring during the time interval (t, t + ∆t ) is negligible.
This is denoted by 0(∆t).
(iii) Almost one event (arrival or departure) can occur during a small time interval ∆t. The probability
of an arrival during the time interval (t, t + ∆t) is given by:
P1(∆t) = λ ∆t + 0 (∆t)
where λ is a constant and is independent of the total number of arrivals up to time t; ∆t is a
small time interval and 0 (∆t) represents the quantity that becomes negligible when compared to
∆t as ∆t → 0, i.e.
lim { 0 ( ∆t )/∆t} = 0
∆ t →0

16.4.1 Distribution of Arrivals (Pure Birth Process)

The arrival process assumes that the customers arrive at the queuing system and never leave it. Such an arrival
process is called pure birth process. The terms commonly used developing of various queuing models are
the following:
∆t = probability of more than one customer’s arrival is negligible, during any given small interval
of time ∆t.
λ∆t = probability that a customer will arrive in the system during time ∆t.
1 – λ∆t = probability that no customer will arrive in the system during time ∆t.
 Queuing Theory 569

Remark If the arrivals are random, then the probability distribution of a number of arrivals in a fixed time
interval follows a Poisson distribution [See Appendix 16.A for proof].

16.4.2 Distribution of Interarrival Times

If the number of arrivals, r in time t follows the Poisson distribution, then

e −λt (λt )r
P(x = r) = , r = 0, 1, 2, . . .
r!

and the inter-arrival time in a fixed period follows the exponential distribution P(x = t) = λ e − λt [ See
Appendix 16.A for proof ].

Markovian property of interarrival times The Markovian property of interarrival times states that the
probability of service of a customer is completed at a particular time, t is independent of time of service.
That is:

Prob { T ≥ t1 | T ≥ t 0 } = Prob { 0 ≤ T ≤ t1 − t 0 }

where T is the time between successive arrivals [ See Appendix 16.A for proof ].

16.4.3 Distribution of Departures (Pure Death Process)

The departure process assumes that no customer arrives in the system while the service is continued for
customers already in the system. Let, there be N ≥ 1 customers in the system at time t = 0 (starting time).
Since the service is being provided at the rate of µ, therefore, the customers leave the system at the rate
µ, after being served. Such a process is called pure death process [See Appendix 16.A for proof ].
Basic Axioms
(i) Probability of the departure during time ∆ t is µ ∆ t.
(ii) Probability of more than one departure between time t and t + ∆t is negligible.
(iii) The number of departures in non-overlapping intervals are statistically independent.
The following terms are used in developing of various queuing models.
µ∆t = Probability that a customer in service at time t will complete service during time ∆t.
1 – µ∆t = Probability that the customer in service at time t will not complete service during time ∆t.

16.4.4 Distribution of Service Times

The probability density function f (t) of service time is given by:

f (t ) =
R|Sµ e − µt
; 0≤t ≤∞
T|0 ; t<0

This shows that service times follows negative exponential distribution with mean 1/µ and variance 1/µ2.
As the service time increases, the probability of such service times tails off exponentially towards zero.
The area under the negative exponential distribution curve is determined as:

F (T ) = z t
0
µ e − µ t dt = − µ e − µ t
−µ t 0 − µt
t
0

= −e +e = 1− e
It is also described as:
F (T ) = f (t ≤ T ) = 1 – e – µ t
where F (T ) is the area under the curve to the left of T. Thus,
1 – F (T ) = f (t ≥ T ) = e – µ t
is the area under the curve to the right of T.
570 Operations Research: Theory and Applications

CONCEPTUAL QUESTIONS A
1. Discuss the fields of application for queuing theory. Explain 9. What is queuing theory? What types of questions are sought to
queue discipline and its various forms. be answered in analysing a queuing system?
2. Explain the basic queuing process. What are the important [Delhi Univ., MBA (HCA), 2004]
random variates in queuing system to be investigated? 10. What do you understand by a queue? Give some important
3. What do you understand by (i) queue discipline, (ii) arrival applications of queuing theory? [Delhi Univ., MBA, 2005]
process (iii) service process? 11. (a) Define a waiting line. Give a brief description of the various
4. Explain, in brief, the main characteristics of the ‘queuing system’. types of queues. [C.C.S. Univ., MBA, 2000]
[Delhi Univ., MBA, 2000] (b) List various possible configurations of the service facilities
5. What is traffic intensity? If traffic intensity is 0.30, what is the in a queuing system.
percentage of time a system remains idle? 12. (a) Give two examples to illustrate the applications of queuing
6. (a) Show that n, the number of arrivals in a queue in time t follows theory in business and industry.
the Poisson distribution, stating the assumptions clearly. [Delhi Univ., MBA, 2002]
(b) Show that the distribution of the number of births up to time (b) In what kind of situations can queuing theory be applied
T in a simple birth process follows the Poisson law. successfully? Give appropriate examples.
7. (a) If the number of arrivals in a particular time interval follows [Delhi Univ., MBA, 2005]
a Poisson distribution, show that the distribution of the time 13. (a) Discuss the essential features of queuing system. What
interval between two successive arrivals is exponential. are the important random variates in a queuing system to
(b) Show that if the interarrival times are negative exponentially be studied? [Delhi, Univ., MBA, 2006]
distributed, then the number of arrivals in a time period is (b) What are some of the operating characteristics of a
a Poisson process, and conversely. queuing system? How can they be used in the evaluation,
8. (a) State and prove the Markovian property of interarrival or design system?
times. 14. Queuing theory can be used effectively in determining optimal
(b) Establish the probability distribution formula for pure-death service levels. Elucidate this statement with the help of an
process. If the intervals between successive arrivals are example. [Delhi Univ., MBA (HCA), 2006]
random variables that follow the negative exponential 15. What is a queuing theory problem? Describe the advantages of
distribution with mean 1/λ then show that the arrivals form queuing theory to a business executive with a view of persuad-
a Poisson distribution with mean λ. ing him to make use of the same in management.

16.5 CLASSIFICATION OF QUEUING MODELS

Queuing theory models are classified by using special (or standard) notations described initially by D.G.
Kendall in the form (a/b/c). Later A.M. Lee added the symbols d and c to the Kendall’s notation. In the
literature of queuing theory, the standard format used to describe queuing models is as follows:
{(a / b / c) : (d / c)}
where a = arrivals distribution
b = service time distribution
c = number of servers (service channels)
d = capacity of the system (queue plus service)
e = queue (or service) discipline
In place of notation a and b, the following descriptive notations are used for the arrival and service times
distribution:
M = Markovian (or Exponential) interarrival time or service-time distribution.
D = Deterministic (or constant) interarrival time or service time.
G = General distribution of service time, i.e. no assumption is made about the type of distribution
with mean and variance.
GI = General probability distribution – normal or uniform for inter-arrival time.
Ek = Erlang-k distribution for interarrival or service time with parameter k (i.e. if k = 1, Erlang is
equivalent to exponential and if k = ∞ , Erlang is equivalent to deterministic).
For example, a queuing system in which the number of arrivals is described by a Poisson probability
distribution, the service time is described by an exponential distribution, and there is a single server, would
be represented by M/M/1.

16.5.1 Solution of Queuing Models

Two approaches – mathematical and simulation, are available for solving queuing models. In this chapter,
the mathematical approach has been discussed where arrival and service time distributions are approximated
by known mathematical distributions, i.e. Poisson, exponential, etc., to derive expected value of performance
 Queuing Theory 571

measures of the queuing system. Further, if the cost associated with the performance measures is also
known, then optimal results can also be determined.
The simultaneous occurrence of arrivals and departures is also called Birth-and-Death Process. This
process helps in determining the probability distribution of the number of customers in the queuing system
at a particular period of time. The probability distribution, so obtained, is then used to determine expected
value of performance measures of queuing models.

16.6 SINGLE-SERVER QUEUING MODELS

Model I: {(M/M/1): ( ∞ /FCFS)} Exponential Service – Unlimited Queue
This model is based on certain assumptions about the queuing system:
(i) Arrivals are described by Poisson probability distribution and come from an infinite calling The simultaneous
population. occurrence of
(ii) Single waiting line and each arrival waits to be served regardless of the length of the queue (i.e. arrivals and
no limit on queue length – infinite capacity) and that there is no balking or reneging. departures is also
called Birth-and-
(iii) Queue discipline is ‘first-come, first-served’. Death Process
(iv) Single server or channel and service times follow exponential distribution.
(v) Customers arrival is independent but the arrival rate (average number of arrivals) does not change
over time.
(vi) The average service rate is more than the average arrival rate.
The following events (possibilities) may occur during a small interval of time, ∆t just before time t.
1. The system is in state n (number of customers) at time t and no arrival and no departure.
2. The system is in state n + 1 (number of customers) and no arrival and one departure.
3. The system is in state n – 1 (number of customers) and one arrival and no departure.
Figure 16.7 illustrates the process of determining Pn (probability of n customers in the system at time
t) when customers are either waiting or receiving service at each state. Customers may arrival or left by
the completion of the leading customer’s service.

Fig. 16.7
Single Serve
Queuing System
States

Step 1: Obtain system of differential difference equations

If Pn (t) is the probability of n customers at time t in the system, then the probability that the system will
contain n customers at time (t + ∆t ) can be expressed as the sum of the joint probabilities of the three
mutually exclusive and collectively exhaustive cases as mentioned above. That is,
For n ≥ 1 and t ≥ 0,
Pn (t + ∆t ) = Pn (t) . Prob (no arrival in ∆t and on departure in ∆t )
+ Pn + 1 (t). Prob (no arrival in ∆t and one departure in ∆t )
+ Pn – 1 (t). Prob (one arrival in ∆t and no departure in ∆t )
= Pn (t) {Prob (no arrival in ∆t) × Prob (no departure in ∆ t )}
Pn + 1 (t) {Prob (no arrival in ∆t) × Prob (one departure in ∆ t )}
+ Pn – 1 (t) {Prob (one arrival in ∆t) × Prob (no departure in ∆ t )}
= Pn(t) {1 – λ ∆t} {1 – µ∆t} + Pn +1(t){1 – λ ∆t}µ ∆t + Pn – 1(t){λ ∆t} {1– µ∆t}
= Pn(t) {1 – (λ + µ)∆t} + Pn +1 (t)µ ∆t + Pn – 1(t) λ ∆t + terms involving (∆t)2
Since ∆t is very small, therefore, terms involving (∆t)2 can be neglected. Subtracting Pn (t) from both sides
and dividing by ∆t, we get:
572 Operations Research: Theory and Applications

Pn (t + ∆t ) − Pn (t )
= λPn −1 (t ) + µPn + 1 (t ) – (λ + µ ) Pn (t )
∆t
Taking limit on both sides as ∆t → 0, the above equation reduces to
d
Pn′ (t ) or {Pn (t )} = λ Pn − 1 (t ) + µ Pn + 1 (t ) − ( λ + µ ) Pn (t ); n ≥ 1 (1)
dt
Similarly, for n = 0 at t ≥ 0 (no customer in the system at time (t + ∆t), and no service completion during
∆t), the resulting equation is:
P0 (t + ∆t ) − P0 (t )
= µ P1 (t ) − λ P0 (t )
∆t
Taking limit on both sides as ∆t → 0, we get:
d
P0′(t ) or {P0 (t )} = µ P1 (t ) − λP0 (t ); n=0 (2)
dt
Step 2: Obtain the system of steady-state equations
In the steady-state, Pn (t) is independent of time t, so that
lim Pn (t ) = Pn
t→∞

and lim
t→∞
d
mP (t )
dt n
r = 0; n = 0 , 1, 2 , . . .

Consequently, Eqs (1) and (2) may be written as:

λ Pn − 1 + µ Pn + 1 − ( λ + µ ) Pn = 0; n≥1 (3)
µP1 − λ P0 = 0; n = 0 (4)
Equations (3) and (4) are referred to as steady-state difference equations for a queuing system. The solutions
of these equations can be obtained either by (i) iterative method, (ii) generating functions method, or (iii) linear
operator method. In this chapter, iterative method has been discussed to find the values of P1, P2, . . . in terms
of P0, λ and µ.
Step 3: Solve the system of difference equations
From Eq. (4), we get:

λP 0 = µP1 or P1 =
FG λ IJ P n=0
0;
H µK
Putting n = 1 in Eq. (3), we get
0 = λ P0 + µ P2 − (λ + µ) P1 or µ P2 = (λ + µ) P1 − λP0

F λ + µ IJ P − λ P = FG λ + µ IJ λ P − λ P = FG λ IJ
= G
2
P0
or P2
H µ K µ 1
H µ Kµ0
µ H µK 0 0

F λ + µ IJ P − FG λ IJ P
= G
P3
H µ K H µK 2 1

F λ + µ IJ FG λ IJ P − FG λ IJ FG λ IJ P = FG λ IJ P ; n = 2
= G
2 3

H µ K H µK H µK H µK
0
H µK 0 0

In general, by using the inductive principle, we get:

Pn = probability of being in state n (i.e. n customers in the system)
F λI
= G J
n
P0 ;
H µK n = 1, 2, . . .
∞
Since ∑ Pn = P0 + P1 + . . . + Pn + . . . = 1 (sum of all probabilities)
n=0

or ∑
∞ FG λ IJ n
P0 = 1 or P0 =
1 λ
<1
n=0 H µK ∞
∑
FG λ IJ n
;
µ
n=0 H µK
 Queuing Theory 573

The denominator of this expression is an infinite geometric series whose sum is:

∑
∞ FG λ IJ n
=
1
n=0 H µK 1−
FG λ IJ
H µK
λ
and hence P0 = 1 – =1–ρ
µ

and Pn =
FG λ IJ FG 1 − λ IJ
n
= ρ n (1 − ρ ); ρ < 1, n = 0 , 1, 2 , . . .
H µK H µK
Step 4: Obtain probability density function of waiting time excluding
service time distribution
The waiting time distribution of each customer in the steady-state condition is equal. Let w be the time required
by the server to serve all the customers present in the system, at a particular time in the steady state condition.
Let φw (t) be the probability distribution function of w, i.e. φw (t) = P(w ≤ t), 0 ≤ t ≤ ∞.
Let s1, s2, . . ., sn be the service time taken by the server to serve each of n customers, respectively. Thus,
R|0 , n=0
w = S| ∑ n
si , n ≥1
T
i =1

The waiting time distribution function for a customer in the system is given by:
 P0 = 1 − ρ , n = 0, t = 0

P{w ≤ t} =    n 

 P  ∑ si ≤ t  , n ≥ 1, t > 0
 i = 1 
Since the service time for each customer is independent and identically distributed, therefore, its
probability density function is, ψs (t) = µe – µt, t > 0, where µ is the mean service rate. Thus,
n
φn (t) = ∑ [Pn × Prob {(n – 1) customers got service at time t}]
i =1
× Prob {one customer is under service during time ∆t}
FG 1 − λ IJ FG λ IJ LM ( µt ) e OP µ ∆t
= ∑
n
n n − 1 − µt

H µ K H µ K MN ( n − 1)! PQ
i =1
Thus, the expression for φ (t) may be written as:
w
R|1 − ρ ; t = 0
φ (t) = P{w ≤ t} = S
z
t
∞
w
|T ∑ P φ (t ) dt n =1
n
0
n

1 − ρ ; t =0
 t
=  ∞
(µt )n −1 ⋅µe −µt dt λ
 ∑ ρ (1 − ρ) ∫
n
; t > 0 and ρ =
n = 1 (n − 1)! µ
0

R|1 − ρ ; t = 0
= S t (µ t ) n − 1
|Tz ρ(1 − ρ) µe
∞
−µt
dt ∑ ;t >0
0 n =1 (n − 1)
R|1 − ρ ; t = 0
= S t
|Tρ(1 − ρ)z µ e 0
− µ (1 − ρ) t
dt

This shows that the waiting time distribution is discontinuous at t = 0 and continuous in the range
0 < t < ∞. Thus, the expression for φw (t) may also be written as:
574 Operations Research: Theory and Applications

φ w ( 0) = 1 – ρ ; t = 0

φ ′w ( t ) = λ (1 – ρ) e– µ (1 – ρ) tdt = λ 1 −
FG IJ
λ −(µ − λ)t
e dt ; t > 0
H µ K
Step 5: Calculate busy period distribution
Let w be the random variable denoting total time (waiting and service) spent by a customer in the system.
Then, the probability density function for the distribution is given by:
d
φ (w) = {φ w (t )}, for ω > 0
dt

λ 1−
FG IJ
λ − (µ − λ )t

=
µ H K
e

z GH IJK
F
∞
λ − (µ − λ )t
λ 1− e dt
0 µ
F λI
λ G1 − J e −(µ − λ)t

=
H µK = (µ – λ) e– (µ – λ)t ; t = 0
λ
µ
Thus, busy period distribution becomes:

z z
∞ ∞
φ( w : w > 0 ) dt = ( µ − λ ) e − ( µ − λ ) t dt = 1
0 0

Performance Measures for Model I

1. (a) Expected number of customers in the system (customers in the line plus the customer being served):
∞ ∞
Ls = Σ n Pn = Σ n (1 − ρ) ρ n , 0 < ρ < 1
n=0 n=0

∞ ∞
= (1 − ρ) Σ nρn = ρ (1 – ρ) Σ nρn – 1
n=0 n=1
2
= ρ (1 – ρ) {1 + 2ρ + 3ρ + . . .}

= (1 − ρ )
|RS ρ |UV [sum of an arithmatico-geometric series]
|T (1 − ρ) 2 |W
ρ λ λ
i.e., Ls = = ; ρ=
1− ρ µ−λ µ
(b) Expected number of customers waiting in the queue (i.e. queue length):
∞ ∞ ∞
Lq = ∑ ( n − 1) Pn = ∑ nPn − ∑ Pn
n =1 n =1 n =1

∞
= ∑ nPn −
LM ∑
∞
Pn − P0
OP = Ls − (1 − P0 )
n=0 N
n=0 Q
λ λ λ2 λ
i.e., Lq = − = ; 1 − P0 =
λ−µ µ µ(µ − λ) µ
2. (a) Expected waiting time for a customer in the queue:

z RS d φ (t )UV dt
∞ ∞
Wq =
0
t⋅
T dt Ww = z t ⋅ λ(1 − ρ)e
0
− (µ − λ )t
dt
 Queuing Theory 575

Integrating by parts, Wq
L te
= λ (1 – ρ) M −
e OP
− µ (1 − ρ) t − µ (1 − ρ) t
∞

N − µ (1 − ρ) µ (1 − ρ) Q 2 2
0

F λ I 1 = λ or
= λ G1 − J
Lq
i.e., Wq
H µ K ( µ − λ ) µ (µ − λ ) 2 λ
(b) Expected waiting time for a customer in the system (waiting and service):
Ws = Expected waiting time in queue + Expected service time
1 λ 1 1 L
i.e., Ws = Wq + = + = or s
µ µ (µ − λ ) µ µ − λ λ
3. The variance (fluctuation) of queue length
∞ F∑ ∞ I 2
Var (n) = ∑ n 2 Pn −
n =1
GH n =1
nPn JK
∞
n 2 Pn − ( Ls )2
∞
= ∑ n (1 − ρ) ρ − G
2 F ρ IJ n
2
= ∑
n =1 n =1 H 1 − ρK
= (1 – ρ) 1⋅ ρ 2 + 2 2 2 2 F ρ IJ
⋅ρ + 3 ⋅ρ + . . . − G 3
2

H 1− ρK
ρ λµ
i.e., Var (n) = 2
=
(1 − ρ ) (µ − λ)2

4. Probability that the queue is non-empty:

P (n > 1) = 1 – P0 – P1

= 1− 1−
FG λIJ FG
− 1−
λ IJ FG λ IJ = FG λ IJ 2

H µ K H µ K H µK H µK
5. Probability that the number of customers, n in the system exceeds a given number k:
∞ ∞
k
P (n ≥ k) = ∑ Pk = ∑ (1 − ρ) ρ
n=k n=k
∞
= (1 – ρ) ρk ∑ ρ n − k
n=k

(1 − ρ ) ρ k
= (1 – ρ) ρk [ 1 + ρ + ρ2 + . . .] = = ρk
(1 − ρ )

F λI
P (n ≥ k) = G J
k
F λI
and P (n > k) = G J
k +1

H µK H µK
6. Expected length of non-empty queue:
Expected length of waiting line
L =
Prob (n > 1)
Lq λ2/µ ( µ − λ ) µ
= = 2 =
P ( n > 1) ( λ /µ ) µ−λ
Example 16.1 A television repairman finds that the time spent on his jobs has an exponential
distribution with a mean of 30 minutes. If he repairs the sets in the order in which they came in, and if the
arrival of sets follows a Poisson distribution with an approximate average rate of 10 per 8-hour day, what
is the repairman’s expected idle time each day? How many jobs are ahead of the average set just brought
in? [Rajasthan, MCom, 2000; Delhi Univ., MCom 2002]
576 Operations Research: Theory and Applications

Solution From the data of the problem, we have:

λ = 10/8 = 5/4 sets per hour; and µ = (1/30) 60 = 2 sets per hour
(a) Expected idle time of repairman each day
Since number of hours for which the repairman remains busy in an 8-hour day (traffic intensity) is given
by:
(8) (λ/µ) = (8) (5/8) = 5 hours
Therefore, the idle time for a repairman in an 8-hour day will be: (8 – 5) = 3 hours
(b) Expected (or average) number of TV sets in the system
λ 5/ 4 5
Ls = = = = 2 (approx.) TV sets
µ−λ 2 − (5 / 4) 3
Example 16.2 In a railway marshalling yard, goods trains arrive at a rate of 30 trains per day. Assuming
that the interarrival time follows an exponential distribution and the service time (the time taken to hump
a train) distribution is also exponential with an average of 36 minutes. Calculate:
(a) expected queue size (line length)
(b) probability that the queue size exceeds 10
If the input of trains increases to an average of 33 per day, what will be the change in (i) and (ii)?
Solution From the data of the problem, we have
λ = 30/60 × 24 = 1/48 trains per minute and µ = 1/36 trains per minute.
The traffic intensity then is, ρ = λ /µ = 36/48 = 0.75
(a) Expected queue size (line length):
ρ 0.75
Ls = = = 3 trains
1− ρ 1 − 0.75
(b) Probability that the queue size exceeds 10:
P (n ≥ 10) = ρ10 = (0.75)10 = 0.06
If the input increases to 33 trains per day, then we have λ = 33/60 × 24 = 11/480 trains per minute and
µ = 1/36 trains per minute.
λ FG 11 IJ (36) = 0.83
Thus, traffic intensity, ρ =
µ
=
H 480K
Hence, recalculating the values for (i) and (ii)
ρ 0.83
Ls = = = 5 trains (approx.), and
1− ρ 1 − 0.83

P (n ≥ 10) = ρ10 = (0.83)10 = 0.2 (approx.)

Example 16.3 Arrivals at telephone booth are considered to be Poisson with an average time of 10
minutes between one arrival and the next. The length of phone calls is assumed to be distributed
exponentially, with a mean of 3 minutes.
(a) What is the probability that a person arriving at the booth will have to wait?
(b) The telephone department will install a second booth when convinced that an arrival would expect
waiting for at least 3 minutes for a phone call. By how much should the flow of arrivals increase in
order to justify a second booth?
(c) What is the average length of the queue that forms from time to time?
(d) What is the probability that it will take a customer more than 10 minutes altogether to wait for the phone
and complete his call?

Solution From the data of the problem, we have

λ = 1/10 = 0.10 person per minute and µ = 1/3 = 0.33 person per minute
(a) Probability that a person has to wait at the booth.
P ( n > 0) = 1 − P0 = λ /µ = 010
. / 0.33 = 0.3
 Queuing Theory 577

(b) The installation of second booth will be justified only if the arrival rate is more than the waiting time.
Let λ ′ be the increased arrival rate. Then the expected waiting time in the queue will be
λ′
Wq =
µ (µ − λ ′ )
λ′
3= or λ ′ = 0.16
0.33 (0.33 − λ ′)
where Wq = 3 (given) and λ = λ ′ (say) for second booth. Hence, the increase in the arrival rate is
0.16 – 0.10 = 0.06 arrivals per minute.
(c) Average length of non-empty queue:
µ 0.33
L= = = 2 customers (approx.)
µ−λ 0.23
(d) Probability of waiting for 10 minutes or more is given by:

P ( t ≥ 10) = z ∞ λ

10 µ
(µ − λ ) e − ( µ − λ ) t dt

z LM e OP
− 0 . 23 t ∞
∞
= (0.3) ( 0.23) e − 0.23t dt = 0.069 = 0 .03
10
N − 0.23 Q 10
This shows that on an average 3 per cent of the arrivals will have to wait for 10 minutes or more before
they can use the phone.
Example 16.4 A warehouse has only one loading dock manned by a three person crew. Trucks arrive
at the loading dock at an average rate of 4 trucks per hour and the arrival rate is Poisson distributed. The
loading of a truck takes 10 minutes on an average and can be assumed to be exponentially distributed. The
operating cost of a truck is Rs 20 per hour and the members of the loading crew are paid Rs 6 each per
hour. Would you advise the truck owner to add another crew of three persons?
Solution From the data of the problem, we have λ = 4 per hour, and µ = 6 per hour.
For Existing Crew
Total hourly cost = Loading crew cost + Cost of waiting time
= {(Number of loaders) × (Hourly wage rate)}
+ {(Expected waiting time per truck, Ws) (Expected arrival per hour, λ)
× (Hourly waiting cost)}
1
= 6×3 + × 4 × 20 = Rs 58 per hour.
6−4
After Proposed Crew
1
Total hourly cost = 6 × 6 + × 4 × 20 = Rs 46 per hour.
12 − 4
Since the total hourly cost after the addition of another crew of three persons is less than the existing
cost, therefore the truck owner must add a crew of another 3 loaders.
Example 16.5 A road transport company has one reservation clerk on duty at a time. He handles
information of bus schedules and makes reservations. Customers arrive at a rate of 8 per hour and the clerk
can, on an average, service 12 customers per hour. After stating your assumptions, answer the following:
(a) What is the average number of customers waiting for the service of the clerk?
(b) What is the average time a customer has to wait before being served?
(c) The management is contemplating to install a computer system for handling information and reservations.
This is expected to reduce the service time from 5 to 3 minutes. The additional cost of having the new
system works out to Rs 50 per day. If the cost of goodwill of having to wait is estimated to be 12 paise,
per minute spent waiting, before being served, should the company install the computer system?
Assume an 8 hours working day.
Solution It is given that λ = 8 per hour; µ = 12 per hour
(a) The average number of customers waiting for the service in the system are
λ 8
Ls = = = 2 customers
µ−λ 12 − 8
578 Operations Research: Theory and Applications

(b) The average time spent by a customer in the system before being served is
1 1 1
Ws = = = hour = 15 minutes.
µ−λ 12 − 8 4
The average waiting time for a customer in the queue is:
λ 8 1
Wq = = = hour = 10 minutes.
µ (µ − λ ) 12 (12 − 8) 6
(c) The additional cost of Rs 50 per day should be compared with the difference in the goodwill cost
of customers with one existing system and with a computer system, before installing the computer system.
The average cost for a customer’s waiting time (W5 = 15 minutes) in the system, is 0.12 × 15 = Rs 1.80.
Also there are 8 arrivals per hour or in 8 hours 64 (= 8 × 8) customers request service at a total goodwill cost
of 1.80 × 64 = Rs 115.20.
By installing a computer system, the computer will increase a clerk’s service rate up to µ = 20 customers
per hour (3 customers per minute). Thus, the average time spent by a customer in the system would be
1 1 1
Ws = = = hour = 5 minutes
µ − λ 20 − 8 12
and the average daily customer queuing (or goodwill) cost would be reduced to: 64 × (0.12 × 5) = Rs 38.40.
An additional cost of having the computer would be Rs 50 per day. Thus, the average total daily cost
would be
TC = Computer cost + Goodwill cost = 50 + 38.40 = Rs 88.40
This cost is less than the existing goodwill loss cost and gives a net saving of Rs (115.20 – 88.40) =
Rs 26.80. Hence, company can install a computer.

Model II: {(M/M/1) : ( ∞ /SIRO)}

This model is identical to the Model I with the only difference in queue discipline. Since the derivation of
Pn is independent of any specific queue discipline, therefore even in this model we have:
Pn = (1 − ρ ) ρ n ; n = 1, 2 , . . .
Consequently, other results will also remain unchanged as long as Pn remains unchanged.

Model III: {(M/M/1) : (N/FCFS)} Exponential Service – Finite (or Limited) Queue
This model is also based on all assumptions of Model I, expect a limit on the capacity of the system to
accommodate only N customers. This implies that once the line reaches its maximum length of N customers,
no additional customer will be allowed to enter into the system.
A finite queue may arise due to physical constraint such as emergency room in a hospital; one-man
barber shop with certain number of chairs for waiting customers, etc.
The difference equations derived in Model I will also be the same for this model as long as n < N.
The system of steady-state difference equations for this model are:
λ P0 = µP1 ;n=0
( λ + µ) Pn = λ Pn − 1 + µ Pn +1 ; n = 1, 2, . . ., N – 1
λ PN −1 = µ PN ;n=N

In this case the service rate does not have to exceed arrival rate (µ > λ) in order to obtain steady-
state conditions.
Using the usual procedure, from the first two difference equations, the probability of a customer in the
system for n = 0, 1, 2, . . ., N is obtained as follows:

F λI
= G J
n
P0 ; n≤ N
Pn
H µK
To obtain the value of P0, use the fact that:
N N
1= ∑ Pn = ∑ ( λ /µ ) n P 0
n=0 n=0
 Queuing Theory 579

N N
= P0 ∑ ( λ /µ ) n = P 0 ∑ ρn
n=0 n=0

= P0 [1 + ρ + ρ 2 + . . . + ρ N ] = P0
LM1 − ρN +1 OP
MN 1 − ρ PQ
1− ρ λ
and consequently, P0 = ; ρ ≠ 1 and ρ= (< 1)
1 − ρ N +1 µ
 1 − ρ  n λ
 N +1 
ρ ; n ≤ N; ≠ 1
 1− ρ  µ
Pn =  
 1 λ
N +1 ; =1
 µ

The steady-state solution in this case exists even for ρ > 1. This is due to the limited capacity of the
system. If λ < µ and N → ∞ , then Pn = ( 1 – λ/µ) (λ/µ)n , which is the same as in Model I.
Performance Measures for Model III
1. Expected number of customers in the system:
N N F 1− ρ I ρ
Ls = ∑
n =1
nPn = ∑
n =1
n GH 1 − ρ JKN +1
n

1− ρ N 1− ρ
= ∑ nρ n = ( ρ + 2 ρ 2 + 3ρ 3 + . . . + Nρ N )
1 − ρ N +1 n = 0 1− ρ
N +1

R| ρ − ( N + 1) ρ N +1

i.e., Ls = S
|1− ρ 1− ρ N +1
; ρ ≠ 1(λ ≠ µ)

|| N ; ρ = 1(λ = µ)
T2
2. Expected number of customers waiting in the queue:
λ λ ( 1 − PN )
Lq = L s − = Ls −
µ µ
3. Expected waiting time of a customer in the system (waiting + service):
Lq 1 Ls
Ws = + =
λ ( 1 − PN ) µ λ ( 1− PN )
4. Expected waiting time of a customer in the queue:
1 Lq
Wq = Ws − or
µ λ ( 1 − PN )
5. Potential customers lost ( = time for which system is busy):
PN = P0 ρN
Effective arrival rate, λeff = λ (1 – PN )
Effective traffic intensity, ρeff = λe /µ.

Example 16.6 Consider a single server queuing system with Poisson input and exponential service
times. Suppose the mean arrival rate is 3 calling units per hour, the expected service time is 0.25 hour and
the maximum permissible calling units in the system is two. Derive the steady-state probability distribution
of the number of calling units in the system, and then calculate the expected number in the system.
Solution From the data of the problem, we have
λ = 3 units per hour; µ = 4 units per hour, and N = 2
Then traffic intensity, ρ = λ /µ = 3/4 = 0 . 75
580 Operations Research: Theory and Applications

The steady-state probability distribution of the number of n customers (calling units) in the system is:
(1 − ρ) ρn (1 − 0.75) (0.75) n
Pn = = = (0.43) (0.75) n ; ρ ≠1
1 − ρ N +1 1 − (0.75) 2 + 1
(1 − ρ ) 1 − 0.75 0.25
and P0 = N +1
= 2 +1
= = 0.431
1− ρ 1 − ( 0.75) 1 − (0.75) 3
The expected number of calling units in the system is given by:
N 2
Ls = ∑ nPn = ∑ n ( 0.43) ( 0.75) n
n=1 n =1
2
= 0.43 ∑ n ( 0.75) n = 0.43 {( 0.75) + 2 ( 0.75) 2 } = 0.81 .
n =1

CONCEPTUAL QUESTIONS B

1. Derive the difference equations for the queuing model {(M /M /1) expression for average length of busy period. Describe a queue
: (∞/ FCFS)}. How would you proceed to solve the model? model and steady- state equations of M/M/1 queues. What is the
2. In a single server, Poisson arrival and exponential service time probability that at least one unit is present in the system.
queuing system show that the probability Pn of n customers in 6. Show that the average number of units in a (M/M/1) queuing
steady-state satisfies the following equations: system is equal to ρ /(1 – ρ). [Raj. Univ., MPhil, 2000]
λ P0 = µP1 ;n =0 7. Customers arrive at a sales counter in a Poisson fashion with,
(λ + µ) P1 = µP2 ;n =1 mean a arrival rate λ and exponential service times with mean
service rate of µ. Determine (a) average length of non-empty
(λ + µ) Pn = µ Pn + 1 + λ Pn – 1 ;n ≥2
queue, (b) average waiting time of an arrival.
3. Show that for a single service station, Poisson arrivals and 8. (a) For the queuing model {(M/M/1) : (N/FCFS)}, the steady-
exponential service time, the probability that exactly n calling units state probability Pn is given by:
in the queuing system is:
Pn = (1 – ρ)ρn ; n ≥ 0 (1 − ρ)
Pn = ρn ; 0 ≤ n ≤ N,
where, ρ is the traffic intensity. Also find the expected number of 1 − ρN + 1
units in the system. (b) Obtain expressions for P0
4. Define cumulative probability distribution of waiting time for a (c) Obtain expected number of customers in the queue and
customer who has to wait. Also show that in an: {(M/M/1) : system separately
(∞/FCFS)} queue it is given by: 9. Explain {(M/M/1) : (N /FCFS)} system and solve it under steady-
state condition. [Garhwal, MSc (Maths), 2001, AMIE,2005]
(1 − ρ) e −µt (1 − ρ) ; ρ = λ/µ
10. For the single server, finite (or limited) queuing system, find the
5. Define the concept of busy period in queuing theory and its (a) average number of customers in the system, and (b) average
distribution for the system {(M/M/1) : ( ∞ /FCFS)}. Obtain the queue length.

SELF PRACTICE PROBLEMS A

Model I an exponential distribution with a mean of 90 seconds. Find the

average waiting time of a customer. Also determine the average
1. At what average rate must a clerk at a super market work in order number of customers in the system and the average queue length.
to ensure a probability of 0.90 so that the customer will not have
4. The mean rate of arrival of planes at an airport during the peak
to wait longer than 12 minutes? It is assumed that there is only
period is 20 per hour, and the actual number of arrivals in any hour
one counter at which customers arrive in a Poisson fashion at an
follows a Poisson distribution. The airport can land 60 planes per
average rate of 15 per hour. The length of service by the clerk
hour on an average, in good weather, and 30 planes per hour in
has an exponential distribution.
bad weather. The actual number landing in any hour follows a
2. Consider a self-service store with one cashier. Assume Poisson Poisson distribution with these respective averages. When there
arrivals and exponential service times. Suppose that on average is congestion, the planes are forced to fly over the field in the stack
nine customers arrive every 5 minutes and that the cashier can awaiting the landing of other planes that arrived earlier.
serve 10 in 5 minutes. Find: (a) How many planes would be flying over the field in the stack,
(a) average number of customers queuing for service on an average, in good weather conditions and in bad
(b) probability of having more than 10 customers in the system, and weather conditions?
(c) probability that a customer has to queue for more than 2 (b) How long would a plane be in the stack and in the process
minutes. of landing in good and in bad weather.
If the service can be speed up to 12 in 5 minutes by using a
5. A repair shop, attended by a single mechanic, has an average of
different cash register, what will be the effect of this on the
four customers an hour who bring small appliances for repair. The
quantities (a), (b) and (c).
mechanic inspects them for defects and for this he takes six
3. Customers arrive at a box office window, being manned by a minutes on an average. Arrivals are Poisson and service rate has
single individual, according to a Poisson input process with a mean an exponential distribution. You are required to:
rate of 30 per hour. The time required to serve a customer has
 Queuing Theory 581

(a) Find the proportion of time during which there is no customer 13. A maintenance service facility has Poisson arrival rates, negative
in the shop. exponential service times, and operates on a first-come, first-
(b) Find the probability of finding at least one customer in the served queue discipline. Breakdowns occur on an average of
shop. three per day, with a range of zero to eight. The maintenance crew
(c) Calculate is the average number of customers in the system? can service, on an average, six machines per day, with a range
(d) Find the average time spent by a customer in the shop from zero to seven. Find the:
including service. (a) Utilization factor of the service facility
(b) Mean waiting time in the system
6. In a bank, cheques are cashed at a single ‘teller’ counter.
(c) Mean number machines in the system
Customers arrive at the counter in a Poisson manner at an
(d) Mean waiting time of machines in the queue
average rate of 30 customers per hour. The teller takes, on an
(e) Probability of finding 2 machines in the system
average, a minute and a half to cash a cheque. The service time
has been shown to be exponentially distributed. 14. Telephone users arrive at a booth following a Poisson distribution
(a) Calculate the percentage of time the teller is busy. with an average time of 5 minutes between one arrival and the
(b) Calculate the average time a customer is expected to wait. next. The time taken for a telephone call is on an average 3
7. In a tool crib manned by a single assistant, operators arrive at the minutes and it follows an exponential distribution. What is the
tool crib at the rate of 10 per hour. Each operator needs 3 minutes, probability that the booth is busy? How many more booths should
on an average, to be served. Find out the loss of production due be established to reduce the waiting time to less than or equal to
to the time lost in waiting for an operator in a shift of 8 hours, if half of the present waiting time.
the rate of production is 100 units per shift. 15. In a factory, the machines breakdown on an average rate of 10
machines per hours. The idle time cost of a machine is estimated
8. Trucks arrive at a factory for collecting finished goods that are to be Rs 20 per hour. The factory works 8 hours a day. The
supposed to be transported to distant markets. As and when they factory manager is considering 2 mechanics for repairing the
come they are required to join a waiting line and are served on first- machines. The first mechanic A takes about 5 minutes, on an
come, first-served basis. Trucks arrive at the rate of 10 per hour average, to repair a machine and demands wages of Rs 10 per
whereas the loading rate is 15 per hour. It is also given that arrivals hour. The second mechanic B takes about 4 minutes in repairing
are Poisson and loading is exponentially distributed. Transporters a machine and demands wages at the rate of Rs 15 per hour.
have complained that their trucks have to wait for nearly 12 hours Assuming that the rate of machine breakdown is Poisson distributed
at the plant. Examine whether the complaint is justified. Also and the repair rate is exponentially distributed, which of the two
determine the probability that the loaders are idle in the above mechanics should be engaged?
problem.
9. On an average 96 patients per 24 hour-day require the service
of an emergency clinic. Also on an average, a patient requires 10
Model III
minutes of active attention. Assume that the facility can handle
16. If in a period of 2 hours, in a day (8 to 10 a.m.), trains arrive at
only one emergency at a time. Suppose that it costs the clinic Rs
the yard every 20 minutes but the service time continues to remain
100 per patient treated to obtain an average servicing time of 10
36 minutes, then calculate, for this period:
minutes, and that each minute of decrease in this average time
(a) The probability that the yard is empty, and
would cost Rs 10 per patient treated, how much would have to be
(b) The average number of trains in the system, on the assump-
budgeted by the clinic to decrease the average size of the queue
tion that the line capacity of the yard is only limited to 4 trains.
from 4/3 patients to 1/2 patient. [IAS (Main) 1995]
17. At a railway station, only one train is handled at a time. The railway
10. In a service department manned by one server, on an average yard is sufficient only for two trains to wait while the other is given
one customer arrives every 10 minutes. It has been found out that a signal to leave the station. Trains arrive at the station at an
each customer requires 6 minutes to be served. Find out: average rate of 6 per hour and the railway station can handle them
(a) Average queue length on an average of 12 per hour. Assuming Poisson arrivals and
(b) Average time spent in the system exponential service distribution, find the steady-state probabilities
(c) Probability that there would be two customers in the queue. for the various number of trains in the system. Also find the
11. A fertilizer company distributes its products by trucks that are average waiting time of a new train arriving at the yard.
loaded at its only loading station. Both, company trucks and [Delhi Univ., MBA, 2003]
contractor’s trucks are used for this purpose. It was found that 18. Patients arrive at a clinic according to a Poisson distribution at
on an average, every 5 minutes one truck arrived and the average the rate of 30 patients per hour. The waiting room does not
loading time was 3 minutes. Out of these trucks 40 per cent belong accommodate more than 14 patients. The examination time per
to the contractors. Making suitable assumptions, determine: patient is exponential with mean rate of 20 per hour.
(a) The probability that a truck has to wait (i) Find the effective arrival rate at the clinic.
(b) The waiting time of a truck that waits (ii) What is the probability that an arriving patient will not wait?
(c) The expected waiting time of customers’ trucks per day Will he find a vacant seat in the room?
(iii) What is the expected waiting time until a patient is dis-
12. Customers arrive at a one-window drive-in bank according to a
charged from the clinic?
Poisson distribution with mean of 10 per hour. Service time per
customer is exponential with a mean of 5 minutes. The space in 19. Assume that goods trains are coming in a yard at the rate of 30
front of the window, including that for the serviced car, can trains per day and suppose that the interarrival times follow an
accommodate a maximum of 3 cars. The other cars can wait exponential distribution. The service time for each train is as-
outside this space. sumed to be exponential with an average of 36 minutes. If the yard
(a) What is the probability that an arriving customer can drive can admit 9 trains at a time (there being 10 lines, one of which is
directly to the space in front of the window? reserved for shunting purpose). Calculate the probability that the
(b) What is the probability that an arriving customer will have to yard is empty and find the average queue length.
wait outside the indicated space? 20. A petrol station has a single pump and space for not more than
(c) How long is an arriving customer expected to wait before 3 cars (2 waiting, 1 being served). A car arriving when the space
starting service? is filled to capacity goes elsewhere for petrol. Cars arrive accord-
582 Operations Research: Theory and Applications

ing to a Poisson distribution at a mean rate of one every 8 minutes. cannot build another pump.) The rent would be Rs 2,000 per
Their service time has an exponential distribution with a mean of month. The expected net profit from each customer is Rs 2 and
4 minutes. the station is open 10 hours everyday. Would it be profitable to rent
The owner has the opportunity of renting an adjacent piece of the additional space?
land, which would provide space for an additional car to wait. (He

HINTS AND ANSWERS

1. λ = 15/60 = 1/4; µ? 10. λ = 60/10 = 6/hour; µ = 60/3 = 10/hour

P ( waiting ≥ 12 ) = z12
∞ λ
µ
b g
µ − λ e − ( µ − λ ) t dt = 0 .10 ( given )
(a) Lq = 0.9 customers, (b) Ws = 1/8 hour or 15 minutes,
(c) P2 = (λ/µ)2 (1 – λ/µ) = 0.144 or 1.44%
( 3 − 12 ) µ 11. λ = 60/5 = 12/hour; µ = 60/3 = 20/hour
Thus e = 0 .4 µ or 1/µ = 2 . 48 minutes/ service.
2. Case I: λ = 9/5, µ = 10/5; (a) P0 (idle time) = λ / µ = 0.6;
(a) Ls = 9 customers, (b) P(n ≥ 10) = (0.9)10 (b) Ws = 1/8 hour or 7.5 minutes

(c) P ( waiting ≥ 2 ) =
λ
µ
Case II: λ = 9/5, µ = 12/5;
z
2
∞
( µ − λ ) e − ( µ − λ ) t dt = 0 .67
(c) Total waiting time = (Number of trucks per day) × (Per cent
contractors trucks) × Expected waiting time for a truck
40 λ
= (12 × 24) × × = 8.64 hours per day
100 µ (µ − λ )
(a) Ls = 3 customers; (b) P ( n ≥ 10) = ( 0.75) 10
12. λ = 10/ hour; µ = 60/5 = 12/ hour

(c) P ( waiting ≥ 2 ) =
λ
µ z
2
∞
( µ − λ ) e − ( µ − λ ) t dt = 0. 30
(a) P0 + P1 + P2 = (1 – λ/µ) + λ/µ(1 – λ/µ) + (λ/µ)2 (1 – λ/µ)
= 0.42
(b) P ( n ≥ 3) = ( λ / µ ) 3 = (10 / 12 ) 3 = 0. 48
3. λ = 30/60 = 0.5; µ = 60/90 = 0.67
(a) Ws = 4.5 minutes/customer; (b) Ls = 3; (c) Lq = 0.25 (c) Wq = 5/12 hour or 25 minutes
4. λ = 20; µ = 60 (in good weather) and 30 (in bad weather) 13. λ = 3/day , µ = 6/day
(a) ρ = λ/µ = 3/6 or 50% (b) Ws = 1/3 day
(a) Lq =
RS 1/6 ( in good weather) (c) Ls = 1 machine (d) Wq = 1.6 day
T 4 /3 (in bad weather) (e) P2 = (λ / µ) 2 (1 − λ/µ) = 0.125
(b) Lq =S
R 1/40 hour ( in good weather) 14. λ = 12/hour; µ = 20/hour
T 1/10 (in bad weather) (a) Busy period, 1 – P0 = λ /µ = 0.60;
5. λ = 4/hour; µ = 60/6 = 10/hour; ρ = λ/µ = 0.4 (b) Wq = 3/40 hour
(a) P0 = 1 – ρ = 0.6; (b) P (n ≥ 1) = (0.4); Ws = 1/8 hour;
(c) Ls = 2/3; (d) Ws = 1/6 hour or 10 minutes 1 1 1
Ws′ = or = , i.e. µ ′ = 28 / hour
6. λ = 30/hour; µ = 60/(3/2) = 40/hour µ′ − λ µ ′ − 12 16
(a) Busy period = 1 – P0 = λ /µ = 3/4, i.e. teller is busy for 75 Thus, number of booths required to achieve new service rate is
per cent of its time. = 28/20 = 1.40 booths.
(b) Ws = 1/10 hour or 6 minutes 15. Mechanic A: λ = 10/hour; µ = 12/hour
7. λ = 10/hour; µ = 20/hour Total cost = Total wages + Cost of non-productive time
(a) Wq = 1/20 hour; average waiting time per shift is: 8/20 = (Hourly rate × No. of hours )
= 2/3 hour; – (Average no. of machines in the system)
Loss of production due to waiting = (2/5) × (100/8) = 5 units × (Cost of idle machine hour )
8. λ = 10/hour; µ =15/hour × (Number of hours)
(a) Lq = 10/75 hour or 8 minutes; λ
(b) Idle time, P0 = 1 – ρ = 5/15 or 33.33% = 10 × 8 + × 20 × 8
µ−λ
9. λ = 96/day; µ = (24 × 60)/10 = 144/day
(i) Lq = 4/3 patients. But if Lq is changed from 4/3 to 1/2, then 10
= 640 + × 160 = Rs 880
new value of µ will be 12 − 10
1 λ2 ( 96 ) 2 Mechanic B: λ = 10/hour; µ = 15/hour
= = or µ ′ = 192 patients/day
2 µ ′ ( µ ′ − λ ) µ ′ ( µ ′ − 96 ) λ
Total cost = 15 × 8 + × 20 × 8
µ−λ
Thus average rate of treatment required is 1/ µ ′ = 60 × (24/192)
= 75 minutes; Decrease in the average time of treatment required 10
= 120 + × 160 = Rs 440
is (10 – 7.5) = 2.5 minutes. Revised budget per patient = Rs (100 15 − 10
+ 2.5 × 10) = Rs 125. Mechanic B should be employed.
 Queuing Theory 583
3 λ
16. λ = 1/20; µ = 1/36; ρ = 36/20 = 1.8(>1) and N = 4; Ls = ∑ nPn = 0.74; Lq = Ls − = 0.24;
n =1 µ
ρ −1
(a) P0 = N+1
= 0.04; Wq = Lq/λ = 0.04
ρ −1
4 18. λ = 30/60 , µ = 20/60 , ρ = 2/3 and N = 14;
(b) L s = ∑ n Pn = 2 .9 ≡ 3.
n=0 Find P0, Pn and Ws.
17. λ = 6; µ = 12; ρ = λ/µ = 0.5 and N = 3; 19. λ = 30/60 × 24 = 1/48 ; µ = λ/16;
ρ = λ/µ = 0.75; P0 = 0.28, Lq = 1.55
1− ρ
P0 = = 0.53 20. λ = 30/60 × 24 = 1/48 ; µ = 1/36; ρ = 0.75;
1− ρN + 1
P0 = 0.28, Ls = 3 trains.

16.7 MULTI-SERVER QUEUING MODELS

Model IV: {(M/M/s) : ( ∞ /FCFS)} Exponential Service – Unlimited Queue
In this case instead of a single server, there are multiple but identical servers in parallel to provide service to
customers. It is assumed that only one queue is formed and customers are served on a first-come, first-served
basis by any of the servers. The service times are distributed exponentially with an average of µ customers
per unit of time. If there are n customers in the queuing system at any point in time, then the following two
cases may arise:
(i) If n < s, (number of customers in the system is less than the number of servers), then there will be no
queue. However, (s – n) number of servers will not be busy. The combined service rate will then be
µn = nµ.
(ii) If n ≥ s, (number of customers in the system is more than or equal to the number of servers) then
all servers will be busy and the maximum number of customers in the queue will be (n – s). The
combined service rate will be µn = sµ.
As in the Model I, Fig. 16.8 shows the movement of customers among several states

Fig. 16.8
Multi-Server
Queuing System
States

Thus to derive the results for this model, we have :

λ n = λ for all n ≥ 0
nµ; n < s
µn = 
 sµ; n ≥ s
The method of determining probability, Pn of n customers in the queuing system at time t and value of
performance measures is summarized below:
Step 1: Obtain the system of differential-difference equations
Using the same logic as in Model I, we have:
Pn (t + ∆t) = Pn (t) {1 – λ ∆t} {1 – n µ ∆t} + Pn + 1 (t) {1 – λ ∆t} {(n + 1) µ ∆t}
+ Pn – 1 (t) {λ ∆t} {1 – (n – 1) µ ∆t}
= – (λ + n µ) Pn (t) ∆t + (n + 1) µ Pn + 1 (t) ∆t + λPn – 1 (t) ∆t + Pn (t)
+ terms involving (∆t)2 ; 1 ≥ n < s
584 Operations Research: Theory and Applications

Pn (t + ∆t) = Pn (t) {1 – λ ∆t} {1 – µ ∆t} + Pn + 1 (t) {1 – λ ∆t} {s µ ∆t}

+ Pn – 1 (t) λ ∆t {1 – s µ ∆t}
= – (λ + s µ) Pn (t) ∆t + s µ Pn + 1 (t) ∆t + λ Pn – 1 (t) ∆t + Pn (t)
+ terms involving (∆t)2 ; n ≥ s
and P0 (t + ∆t) = P0 (t) (1 – λ ∆t) + P1 (t) µ ∆t; n = 0
By dividing these equations by ∆t and then by taking limit as ∆t → 0, we get
Pn′ ( t ) = – (λ + nµ) Pn (t) + (n + 1) µ Pn + 1 (t) + λ Pn – 1 (t) ; 1≤n<s
Pn′ ( t ) = – (λ + sµ) Pn (t) + s µ Pn + 1 (t) + λ Pn – 1 (t) ; n≥s
and P0′ ( t ) = – λ P0 (t) + µ P1 (t) ; n=0

Step 2: Obtain the system of steady-state equations

In the steady-state condition, the differential-difference equations obtained from the above equations as
t → ∞, are:
– λP0 + µ P1 = 0 ; n=0
– (λ + nµ) Pn + (n + 1) µ Pn + 1 + λPn – 1 = 0 ; 0<n<s
– (λ + sµ) Pn + s µPn + 1 + λPn – 1 = 0 ; n≥s

Step 3: Solve the system of difference equations

Applying the iterative method, the probability of n customers in the system is given by:
 ρn
 P0 ; n≤s
 n!
Pn =  n
 ρ P ; n > s ; ρ = λ / sµ
 s! sn − s 0

Using the following condition to find value of P0
∞ s −1
1 = ∑ Pn = ∑ Pn + ∑ Pn = ∑
∞ FG IJ P + ∑ 1 FG λ IJ P
s −1 1 λ
n
∞
n

n=0 n=0 n=s H K s! s

n=0 H sµ K
n! µ 0
n=s
n−s 0

LM s −1 s F λ I
n n
s FG λ IJ P OP
∞ n n
= P0
MN ∑
n=0
G J
n! H s µ K
+ ∑
s! s
n=s H sµ K PQ n−s 0

=
L
P M ∑
s −1 ( sρ ) n
+
s s O L ( sρ) + s ρ OP ; ρ = λ
∞
∑ ρ P = P M ∑ n
s −1 n s s
0
MN n=0 n! s! PQ MN
n=s n! s! 1 − ρ PQ 0
sµ
n=0
∞
[Since ∑ ρ n = ρs + ρs + 1 + . . . = ρs/(1 – ρ), sum of infinite G.P. ; ρ < 1]
n=s
Thus the probability that the system shall be idle is:

P0 =
LM ∑
s −1 ( sρ ) n 1 ( sρ ) s
+
OP −1
; ρ = λ/sµ
MN
n=0 n! s! 1 − ρ PQ
−1
 s − 1 1  λ n 1  λ  s sµ 
=  ∑   +  
 n = 0 n !  µ 

s !  µ  sµ − λ 


Performance Measures for Model IV

1. The expected number of customers waiting in the queue (length of line):
∞ ∞ ρn
Lq = ∑ ( n − s ) Pn = ∑ ( n − s ) P0
n=s n=s sn − s ⋅ s!
 Queuing Theory 585

ρ s P0 ∞ ρ s P0 ∞ λ
= ∑ (n − s) ρ n − s = ∑ mρ m ; n – s = m, ρ =
s! n=s s! m=0 µ

=
ρs ∞
⋅ ρP0 ∑ mρ m − 1 =
ρs
⋅ ρP0
d LM ∑ ∞
ρm
OP
s! m=0 s! dρ Nm =1 Q
ρs 1 LM F I
1 λ
s
λ ⋅ sµ OP
=
s!
ρ P0
(1 − ρ ) 2
=
MN GH JK
s! µ ( sµ − λ ) 2 PQ P
0

L 1 F λ I λµ OP s
= M
Lq
MN ( s − 1)! GH µ JK ( sµ − λ ) PQ P 2 0

2. The expected number of customers in the system:

λ
Ls = Lq +
µ
3. The expected waiting time of a customer in the queue:

 1  λ s µ  Lq
Wq =    2
 P0 =
 ( s − 1)!  µ  ( sµ − λ)  λ

4. The expected waiting time that a customer spends in the system:

1 Lq 1
Ws = Wq + = +
µ λ µ
5. The probability that all servers are simultaneously busy (utilization factor):

P (n ≥ s) = ∑
∞
Pn = ∑
∞ 1 FG λ IJ n
P0
n=s n=s s! s n − s H µK
1 λ FG IJ s
P0 ∑
∞ FG λ IJ 1 λ sµ
s
=
s! µ H K m=0 H µK =  
s !  µ  sµ − λ
P0

Example 16.7 A super market has two sales girls at the sales counters. If the service time for each customer
is exponential with a mean of 4 minutes, and if the people arrive in a Poisson fashion at the rate of 10 an hour,
then calculate the:
(a) probability that a customer has to wait for being served? [Banasthali, MSc (Maths), 2000]
(b) expected percentage of idle time for each sales girl?
(c) if a customer has to wait, what is the expected length of his waiting time?
[Meerut, MSc (Maths), 2001; Delhi Univ., MBA, 2005]
Solution From the data of the problem, we have
λ = 1/6 per minute; µ = 1/4 per minute, s = 2 and ρ = λ/sµ = 1/3
L
= M Σ
2 −1 1 1
(4/6) n + (4/6) 2
2 ⋅ (1/4) OP −1
FG 2 1 IJ −1
1
Therefore, P0
MN n=0 n! 2! l
2 ⋅ (1/4) − (1/6) q PQ = 1+
H +
3 3 K =
2

and P1 = ( λ /µ ) P0 = ( 4/ 6 )( 1/ 2 ) = ( 1/ 3 )
(a) The probability of having to wait for service:

P(n ≥ 2) =
1 λ
  .
s
sµ
.P0 = 1 4 FG IJ 2
2( 1/4 ) 1 FG IJ 1
s !  µ  sµ − λ 2! 6 HK .
2( 1/4 ) − ( 1/6 ) 2 HK =
6
586 Operations Research: Theory and Applications

(b) The fraction of time the servers are busy, ρ = λ/sµ = 1/3 . Therefore, the expected idle time for each
sales girl is (1 – 1/3) = 2/3 = 67%.
(c) The expected waiting time for a customer in the system:

1 Lq 1 = 1 λ FG IJ s
⋅
µ
⋅ P0 +
1
Ws = Wq +
µ
=
λ
+
µ ( s − 1)! µ H K ( sµ − λ ) 2
µ

FG 4 IJ 2
1/ 4 1
=
H 6K ( 1/ 2 ) − ( 1/ 6 )
2
×
2
+ 4 = 4.5 minutes

Example 16.8 A bank has two tellers working on the savings accounts. The first teller only handles
withdrawals. The second teller only handles deposits. It has been found that the service time distribution for
the deposits and withdrawals, both, are exponential with mean service time 3 minutes per customer. Depositors
are found to arrive in a Poisson fashion throughout the day with a mean arrival rate of 16 per hour. Withdrawers
also arrive in a Poisson fashion with a mean arrival rate of 14 per hour. What would be the effect on the average
waiting time for depositors and withdrawers if each teller could handle both the withdrawals and deposits.
What would be the effect if this could only be accomplished by increasing the service time to 3.5 minutes?
Solution Initially there are two independent queuing systems: Withdrawers and Depositors, where arrivals
follow Poisson distribution and the service time follows exponential distribution.
For Withdrawers Given that, λ = 14/hour and µ = 3/minute or 20/hour
λ 14 7
Average waiting time in queue, Wq = = = hour or 7 minutes
µ ( µ − λ) 20(20 − 14) 60
For Depositors Given that, λ = 16/hour; and µ = 3/minute or 20/hour
λ 16 1
Average waiting time in queue, Wq = = = hour or 12 minutes
µ( µ − λ) 20(20 − 16) 5
Combined Case In this case there will be a common queue with two servers (tellers). Thus, we have
λ = 14 + 16 = 30/hour, µ = 20/hour; s = 2; ρ = λ / sµ = 3/4 .

L 1 F λ I 1 F λ I F sµ I OP n s
−1
L
=M Σ
1 1 3 FG IJ n
1 3FG IJ FG 40 IJ OP
s −1

=M∑
s −1

MN n! GH µ JK + s! GH µ JK GH sµ − λ JK PQ HK H K H 40 − 30K PQ
+
Now, P0
n=0 MN n=0 n! 2 2! 2

L 3 1 F 9I O = 1
= M1 + + G J ⋅ 4P
−1

N 2 2 H 4K Q 7
Average waiting time of arrivals in the queue:
Lq LM 1 F λ I s
µ OP
Wq =
λ
=
MN ( s − 1)! GH µ JK ( sµ − λ ) 2 PQ P 0

FG 3IJ 2
20 1 9
Wq = H 2K (40 − 30) 2
×
7
=
140
hour or 3.86 minutes.

Combined waiting time with increased service time, when λ = 30/hour, µ = 60/3.5 or 120/7 per hour, we
have:
L 1
F I n
F I 2 OP −1
LM OP −1
=M ∑
1 21 1 21 2 .( 120 / 7 ) 7 49 1
P0
MN n=0 H K
n ! 12
+
2 ! 12 H K 2 ( 120 / 7 ) − 30 PQ = 1+
N +
4 4 Q =
15
Average waiting time of arrivals in the queue:
1 λ FG IJ
µ
s
7 FG IJ 2
120/7 1
Wq =
H K
( s − 1)! µ ( sµ − λ ) 2
⋅ P0 =
4 H K ( 120/ 7 − 30 ) 2
×
15
343
= × 60 hour or 11.43 minutes
30
 Queuing Theory 587

Example 16.9 A tax consulting firm has 4 service counters in its office for receiving people who have
problems and complaints about their income, wealth and sales taxes. Arrivals average 80 persons in an
8-hour service day. Each tax adviser spends an irregular amount of time servicing the arrivals, which have
been found to have an exponential distribution. The average service time is 20 minutes. Calculate the
average number of customers in the system, average number of customers waiting to be serviced, average
time a customer spends in the system, and average waiting time for a customer. Calculate how many hours
each week does a tax adviser spend performing his job. What is the probability that a customer has to wait
before he gets service? What is the expected number of idle tax advisers at any specified time?
[IAS (Maths), 1996]
Solution Given that, λ =10/hour; µ = 3/hour, s = 4; and ρ = λ / sµ = 5/ 6
(a) The probability of no customer in the system:

P =M ∑
L 1 F λ I 1 F λ I F sµ I OP
s −1 n s −1

0
MN n! GH µ JK + s ! GH µ JK GH sµ − λ JK PQ
n=0

= M ∑
L 1 F 10 I + 1 F 10I 12 OP
3 n 4 −1

MN n ! GH 3 JK 4! GH 3 JK 12 − 10 PQ
n=0

L 10 1 F 10 I + 1 F 10 I + 1 F 10 I ⋅ 6 OP
= M1 + +
2 3 4 −1

MN 3 2 H 3 K 6 H 3 K 24 H 3 K PQ = 0.021

(b) The average number of customers in the system:

L 1 F λ I λµ OP λ
s
= M
λ
L =L +
s q
µ MN ( s − 1) ! GH µ JK ( sµ − λ ) PQ P + µ
2 0

=M
L 1 F 10 I 30 OP × 0. 021 + 10 = 6.57
4

MN 3 ! H 3 K (12 − 10 ) PQ2 3
(c) The average number of customers waiting in the queue (queue length):
λ
Lq = Ls − = 6 . 57 − (10 / 3) = 3. 24 customers
µ
(d) The average time a customer spends in the system:
1 Lq 1 3. 24 1
|Ws = Wq + = + = + = 0.657 hour or 39.42 minutes
µ λ µ 10 3
(e) The average time a customer waits for service in the queue:
Lq 3.24
Wq = = = 0.324 hour or 19.44 minutes
λ 10
(f ) The time spent by a tax counsellor, i.e. utilization factor:
λ 5
ρ= = = 0.833 hour or 50 minutes
sµ 6
The expected time spent in servicing customers during an 8-hour day is 8 × 0.833 = 6.66 hours. Thus, on
average, a tax advisor is busy for 6.66 × (40/8) = 33.30 hours, based on a 40 hours week.
(g) The probability that a customer has to wait:
s
1  λ  sµ 1 10FG IJ
4
4×3
Pw ( n ≥ s) =  
s !  µ  sµ − λ
⋅ P0 =
4! 3 H K
⋅
4 × 3 − 10
(0.021) = 0.622

(h) The expected number of idle advisers at any specified time can be obtained by adding the probability
of 3 idle, 2 idle and l idle advisers. That is:
Expected number of idle advisers = 4 P0 + 3 P1 + 2 P2 + P3
= 4(0.021) + 3(0.070) + 2(0.118) + 0.131 = 0.661
This means, less than one (0.661) adviser is idle on an average at any instance of time.
588 Operations Research: Theory and Applications

Model V: {(M /M/s) : (N/FCFS)} Exponential Service – Limited (Finite) Queue

This model is an extension of Model IV. However, the assumption of an unlimited waiting area for customers
is not valid in certain cases:
(i) The parking area once full to its capacity, turns away arriving vehicles.
(ii) In a production facility, parts arriving from a previous production stage to a machine for further
processing wait on a conveyer belt, with limited capacity. If the waiting parts fill the belt to its
capacity, the production at the previous stage must come to a halt.
In such a situation, the arriving customers turned away may or may not come back. Hence, the cost
associated with losing a customer should be taken into consideration, along with the cost per server and
the cost of waiting. Let:
λ ; n≤ N
λn = 
0 ; n > N,
nµ ; n < N
and µn = 
 sµ ; s ≤ n ≤ N
The probability of n customers in the system in the steady-state condition is:
 1  λ n
   p0 ; 0≤n≤s
 n!  µ 

Pn =  n
1 λ
 n − s   P0 ; s<n≤ N
 s !s µ
0 ; n>N

N s−1 N
The condition ∑ Pn = ∑ Pn + ∑ Pn = 1
n=0 n=0 n=s

L 1 F λI n
FG λ IJ OP n −1

P = M ∑
s −1 1 N
gives 0
n=0
G
MN n! H µ K J + ∑
s! s H µ K PQ
n=s
n−s

1 F λI FG λ IJ P = 1
s −1 n n
1 N
or ∑
n=0
G J
n! H µ K
P + ∑
s! s
0
H µK
n=s
n−s 0

L s FλI n
FG λ IJ OP = 1 n
P M ∑
s −1 n
s N n
0
n=0
G
MN n! H sµ K J + ∑
s! s H sµ K PQ
n=s
n−s

L ( sρ) + ( sρ) ∑ ρ OP = 1 ; ρ = λ/s µ

P M ∑
s −1 n s N−s
n
0
N n!
n=0 s! Q n=0

N−s 1 − ρ N − s + 1
ρ n  ; ρ (= λ /sµ) ≠ 1, (sum of GP of N − s + 1 terms)
Here ∑ =  1− ρ
n=0
N − s + 1 ; ρ =1

Thus the probability, P0 that the system shall be idle is
L FG IJ sµ R|S1 − FG λ IJ
s N − s +1 U|OP −1

= M Σ
s −1 ( sρ) n 1 λ
P0
MN +
H K sµ − λ |T H sµ K V|P ; ρ = λ /sµ ( = 1)
n=0 n! s! µ
WQ
L s −1
1 F λI
s O −1

= M Σ ( N − s + 1) P
( sρ ) n
MN n=0 n!
+ G J
s! H µ K PQ
Remarks 1. For N → ∞ and λ /s µ < 1 , the above result corresponds to that of Model IV.
2. For s = 1, the above result reduces to the form given in Model III.
 Queuing Theory 589

Performance Measures of Model V

1. The effective arrival rate, λ e = λ (1 − PN )
1. The effective traffic intensity, ρ e = λ e/µ
2. The expected number of customers in the queue
N
Lq = ∑ ( n − s ) Pn = ∑
N
( n − s)
sn FG λ IJ n
P0
n=s n=s s! s n − s H sµ K
N ( sρ ) n
= ∑ (n − s) P0
n=s s! s n − s

( sρ ) s P0 ρ n−s
= ∑ x ⋅ρx − 1 ; x = n – s ; ρ = λ/sµ.
s! x=0

=
( sρ) s ρ P0 N−s
∑
d
(ρ x ) =
( sρ ) s ρ d LM ∑
N −s OP
ρ x P0
s! x=0 dρ s ! dρ MN
x=0 PQ
( sρ ) s ρ d
= 1 + ρ + ρ 2 + . . . + ρ N − s P0
s ! dρ

=
LM
( sρ ) s ρ d 1 − ρ N − s + 1
P0 ,
OP (GP of N – s + 1 terms)
s ! dρ MN
1− ρ PQ
( sρ ) s ρ
Lq = 1 − ρ N − s + 1 − (1 − ρ ) ( N − s + 1) ρ N − s P0
s !(1 − ρ) 2

3. The expected number of customers in the system:

Ls = Lq +
FG λ IJ (1 − P s −1
( s − n) λ FG IJ n

N ) = Lq + s − P0 ∑
H µK n=0 n! µ H K
4. The expected waiting time in the system:
Ls
Ws =
λ (1− PN )
5. The expected waiting time in the queue:
1 Lq
Wq = Ws − =
µ λ ( 1 − PN )
6. The fraction server idle time:
Ls − Lq ρe
1 − = 1 −
s s
Remark If no queue is allowed, then the number of customers who intend to join a queuing system
should not exceed the number of servers, i.e. n ≤ s. Thus, we have:

FG IJ n L s 1 F λ I n OP −1

= M ∑
1 λ
Pn =
n! µ H K P0 and P0
MNn 0 n ! GH µ JK PQ
=

Since no queue forms, Lq = Wq = 0.

The performance measures for this special case of Model V are:
1 λ FG IJ s
P0
1. The fraction of potential customers loss, Ps =
s! µ H K
2. The effective arrival rate, λeff = λ (1 – Ps )
3. The expected waiting time in the system, Ws = 1/µ
590 Operations Research: Theory and Applications

4. The expected number of customers in the system, Ls = λ Ws = λ /µ

ρeff
5. The fraction idle time for server = 1 – .
s
Example 16.10 Let there be an automobile inspection situation with three inspection stalls. Assume that
cars wait in such a way that when a stall becomes vacant, the car at the head of the line pulls up to it.
The station can accommodate almost four cars waiting (seven in station) at one time. The arrival pattern
is Poisson with a mean of one car every minute during the peak hours. The service time is exponential with
a mean of 6 minutes. Find the average number of customers in the system during the peak hours, the
average waiting time and the average number per hour that cannot enter the station because of full capacity.
Solution From the data of the problem, we have:
λ = 1 car per minute; µ = 1/6 car per minute; s = 3, N = 7.
The traffic intensity then becomes ρ = λ/µ = 6. Therefore,

1 LM ∑
3 −1
(6) n + ∑
|| P0 =
7 (6) n OP −1
=
1
n=0 n! MN n=3 3
n −3
3! PQ ,
1141
(a) The expected number of customers in the queue:

Lq =
(3 × 6) 3 ⋅ 6
⋅
FG 1 IJ 1 − (6) 5
− ( −5) (5) (6) 4 = 3.09 cars
3 !( − 5) 2 H 1141
, K
(b) The expected number of customers in the systems:
2 ( 3 − n)
Ls = 3.09 + 3 − P0 ∑ (6) n = 6.06 cars
n=0 n!
(c) The expected waiting time in the system:
6.06 6.06
Ws = = = 12.3 minutes
1(1 − P7 ) (6) 7
F 1 IJ
×G
1−
3 ! 34 H 1,141K
1
Pn =
λ FG IJ
P0 for s < n ≤ N
n

since
s! s n − s µ H K
(d) The expected number of cars per hour that cannot enter the station:

|| 60 λ PN = 60 ⋅ 1⋅ P7 = 60 ⋅ (6)
7
1 FG
= 30.4 cars per hour IJ
3! 34 1,141 H K
CONCEPTUAL QUESTIONS C

1. Obtain the steady-state equations for the model {(M/M/s) : ( ∞ / 2. Obtain the steady-state solution for the model {M/M/s : ∞ /FCFS}
FCFS)}, and show that in the waiting time problem. Obtain the mean queue length, the
 1  λ n average number of customers in the system, and the average
   P0 ; n = 0, 1, 2, …, s waiting time in the system and queue respectively. Explain M used
n !  µ  in this model.
Pn = 
n 3. Obtain the system of steady-state equations and find the value of
 1 λ
 n −s  µ 
P0 ; n = s, s + 1, … P (in usual notations) when (a) n ≤ s and (b) n ≥ s.
s ! s   4. Describe the general problem of M/M/k queuing system and
−1 deduce an explicit expression for the steady-state probability of
 s − 1 1  λ n 1  λ  sµ 
where P0 =  ∑  the length of the queue in an M/M/1 system.
  +  
n = 0 n !  µ  s !  µ  sµ − λ  5. State the basic axioms governing Poisson queues. Find the

distribution of arrivals for the Poisson queues.
[Garhwal, MSc (Maths), 2000]
 Queuing Theory 591

SELF PRACTICE PROBLEMS B

1. Four counters are being opened on the border of a country for 5. A small bank has two tellers, one for deposits and one for
checking the passports and necessary papers of the tourists. The withdrawals. The service time for each teller is exponentially
tourists choose a counter at random. If the arrivals at the border distributed, with a mean of 1 min. Customers arrive at the bank
is Poisson at the rate λ and the service time is exponential with according to a Poisson process, with mean rate 40 per hour; it is
parameters λ/2, what is the steady-state average queue at each assumed that depositors and withdrawers constitute separate
counter? Poisson processes, each with mean rate 20 per hour, and that no
2. A telephone exchange has two long distance operators. The customer is both a depositor and a withdrawer. The bank is
telephone company finds that during the peak load, long distance thinking of changing the current arrangement to allow each teller
calls arrive in a Poisson fashion at an average rate of 15 per hour. to handle both deposits and withdrawals. The bank would expect
The length of service on these calls is approximately exponentially that each teller’s mean service time would increase to 1.2 minutes,
distributed with mean length of 5 minutes. but it hopes that the new arrangement would prevent long lines in
(a) What is the probability that a subscriber will have to wait for front of one teller while the other teller is idle, a situation that occurs
his long distance call during the peak hours of the day? from time to time under the current set-up. Analyze the two
(b) If subscribers wait and are serviced in turn, what is the arrangements with respect to the average idle time of a teller and
expected waiting time? Establish the formulae used. the expected number of customers in the bank at any given time.
3. An insurance company has three claim adjusters in its branch 6. In machine maintenance firm, a mechanic repairs four machines.
office. People with claims against the company are found to arrive The mean time between service requirement is 5 hours for each
in a Poisson fashion, at an average rate of 20 per 8-hour day. The machine and forms an exponential distribution. The mean repair
amount of time that an adjuster spends with a claimant is found machine down time costs Rs 25 per hour and the machine costs
to have exponential distribution with a mean service time of 40 Rs 55 per day of an 8 hour day. (a) Find the expected number of
minutes. Claimants are processed in the order of their appearance. operating machines (b) Determine the expected down time cost
(a) How many hours a week can an adjuster expect to per day (c) Would it be economical to engage two mechanics,
spend with claimants? each repairing only two machines?
(b) How much time, on an average, does a claimant spend in 7. A car servicing station has two bays where service can be offered
the branch office? simultaneously. Due to space limitation, only four cars are ac-
4. A company currently has two tool cribs, each having a single cepted for servicing. The arrival pattern is Poisson with 120 cars
clerk, in its manufacturing area. One tool crib handles only the per day. The service time in both ways is exponentially distributed
tools for the heavy machinery, while the second one handles all with µ = 96 cars per day, per bay. Find the average number of
other tools. It is observed that for each tool crib the arrivals follow cars in the service station, the average number of cars waiting to
a Poisson distribution with a mean of 20 per hour, and the service be serviced, and the average time a car spends in the system.
time distribution is negative exponential with a mean of 2 minutes. 8. A steel fabrication plant is considering the installation of a second
The tool manager feels that if tool cribs are combined in such tool crib in the plant to save walking time of the skilled craftsmen
a way that either of the clerks can handle any kind of tool who check equipment at the tool cribs. The Poisson/exponential
according to the demand, it would prove to be more efficient. He assumptions about arrivals are justified in this case. The time of
also believes that the waiting problem could be reduced to some the craftsmen is valued at Rs 20/hour. The current facility
extent. It is believed that the mean arrival rate at the two tool cribs receives an average of ten calls per hour; with two cribs, each
will be 40 per hour; while the service time will remain unchanged. would average five calls per hour. Currently, there are two
Compare in status of queue and the proposal, with respect to attendants, each of whom services one craftsman per hour. Each
the total expected number of machines at the tool crib(s), the could per form just as well in a separate tool crib. There would be
expected waiting time including the service time for each me- added average inventory costs over the year of Rs 2/hour with
chanic and probability that he has to wait for more than five the separate tool cribs. However, each craftsman would require
minutes. [Delhi Univ., MBA, 2004] six minutes less walking time per call. Evaluate the proposal to set
up a new crib so that each attendant would be able to run one crib.
[Delhi Univ., MBA, 2002]

HINTS AND ANSWERS

1. λ = λ ; µ = λ / 2 , s = 4 and ρ = λ / s µ = 1/ 2 1 λFG IJ s
µ
⋅ P0 = 3.2 minutes
L s −1
( sρ) n
1 ( sρ) s OP −1
Wq =
( s − 1) ! µ H K ( sµ − λ ) 2
P =M ∑
3
+ = ;
0
MN n=0 n! s ! (1 − ρ) PQ 23 3. λ = 20/8 = 5/2 ; µ = 60/40 = 3/2 , s = 3 and ρ = λ/s µ = 5/9
n

Lq =
1 λ FG IJ s
λµ
P0 =
4 P0 = 24/139 ; Pn =
λ1
n−s  µ 
P0 ,
(n − 1) ! µ H K ( sµ − λ ) 2 23
s! s
P1 = 40/139, P2 = 100/147
 

2. λ = 15 /60 = 1/ 4 ; µ = 1/5 , s = 2, and ρ = λ /s µ = 5/8 Expected number of adjusters at any specified time will be

P0 =
LM ∑
s −1
( sρ) n
+
1 ( sρ) s OP −1
=
3
;
3 P0 + 2 P1 + P2 = 4/3
MN
n=0 n! s ! (1 − ρ) PQ 13
Prob (adjuster is idle) = 4/(3 × 3) = 4/9,
Prob (adjuster is busy) = 1 – (4/9) = 5/9,
1 λ FG IJ s
sµ
⋅ P0 = 0.48
(a) (5/9) (8) (5) = 22.2 hours 15 days week,
| P(n = 2) =
s! µ H K sµ − λ (b) Ws = 49 minutes.
592 Operations Research: Theory and Applications

4. (a) When tool crib works independently Ws = P1 + 2P2 = P0 (0.4 + 2 × 2 × 0.08) = 0.38
(α) λ = 20/hour; µ = 2/minute or 30/hour, ρ = λ / µ = 2 /3 (iii) The cost of expected down time per day:
0.38 × 2 × 8 × 25 = Rs 152
ρ (iii) Total cost = 152 + 2(55) = Rs 262
(a) Ls = =2 ; Ws = Ls /λ = 6 minutes (each)
1− ρ Since the total cost with new policy (Rs 262) is more than the
total cost of existing policy (Rs 255), it is uneconomical to
(a) P ( t > 5 ) = e − µ ( 1 − ρ ) t = e − 30 ( 1− 2 / 3 ) × 5/ 60 = e − 0 . 833 = 0 .435 engage two mechanics.
(b) When both tool cribs work jointly 8. λ = 5/hour, µ = 15/minutes or 4/hour, s = 2, N = 4
(a) λ = 40/hour; µ = 30/hour, s = 2 and ρ = λ/sµ 2/3 −1
 s −1 1  λ  n 1 λ
s
sµ   λ  N − s + 1  
P0 =  ∑ 
LM ∑
s −1
( sρ) n 1 ( sρ) s OP −1
1 
  +  
n = 0 n !  µ  s !  µ  sµ − λ
1 −  
  
sµ 
 
(a) P0 = + = ,
MN
n=0 n! s ! (1 − ρ) PQ 5
LM 5 1 F 5 I 2 ( 8) R| F 5 I
2 3−2+1 U| OP −1

L s = Lq +
λ
µ
= 2 .4 persons; Ws =
λ
= 3. 6 minutes.
Ls =
MN1 + 4 + 2 ! GH 4 JK 2 ( 8 ) − 5 ST|1 − GH 8 JK V| P
WQ
= 28

7. λ = 1/5 per hour, µ = 1/hour, s = 4; R| 1 F 5 I 0.28

n

Expected number of machines in the system; P0 = 0.4

Pn =
|S n! GH 4 JK b g ;0≤n≤2

Ws = P1 + 2P2 + 3P3 + 4P4 = P0 (1 × 0.8 + 2 × 0.48 + || 1 FG 5 IJ b 0.28g ; 2 ≤ n ≤ 3

n

3 × 0.192 + 4 × 0.038) = 1 (approx.)

Down time therefore is 8 machine hours per day
T 2! ( 2 ) H 4 K
n−2

s −1 n
(a) The number of operating machines = 4 – Ws or 4 – 1 = 3 ( s − n)  λ 
Ls = Lq + s − P0 ∑  
(b) The cost of expected down time per day (8-hour day); n=0 n! µ 
1 × 8 × 25 = Rs 200
3 2 −1 n
(ii Total cost = Down time cost × Operator cost = 200 + 55 (2 − n)  5 
= ∑ (n − 2) Pn + 2 − (0.28) ∑  
= Rs 255 n=2 n=0 n!  4
(c) When each mechanic is given 2 machines to repair, P0 = 0.68
(iii) The expected number of machines in the system: = 3 cars (approx.)

16.8 FINITE CALLING POPULATION QUEUING MODELS

Model VI: {(M/M/1) : (M/GD)} Single Server – Finite Population (Source) of Arrivals
This model is similar to Model I except that the calling population of potential customers is limited, say
M. Thus, the arrival of additional customers is not allowed to join the system when the system is busy
in serving the existing customers in the queue. Few applications of this model are as under:
(i) A fleet of office cars available for 5 senior executives. Here these 5 executives are the customers,
and the cars in the fleet are the servers.
(ii) A maintenance staff provides repair to M machines in a workshop. Here the M machines are
customers and the repair staff members are the servers.
When there are n customers in the system, then the system is left with the capacity to accommodate
M – n more customers. Thus, further arrival rate of customers to the system will be λ(M – n). That is, for
s = 1, the arrival rate and service rate is stated as follows:

λn =
RS
λ ( M − n ) ; n = 1, 2 , . . ., M
0 T
;n> N
µn = µ ; n = 1, 2 , . . ., M
Performance Measures of Model VI
Substituting for λ n and µ n in the expression for Pn and P0 in Model V, we get:
1. The probability that the system is idle:
L FG IJ n OP −1

=M∑
M M! λ
P0
MNn=0 ( M − n)! µ H K PQ
2. The probability that there are n customers in the system:

Pn =
M! λFG IJ n
P0 ; n = 1, 2 , . . ., M
( M − n) ! µ H K
 Queuing Theory 593

3. The expected number of customers in the queue (or queue length):

M λ+µ FG IJ
Lq = ∑ (n − 1) Pn = M −
n =1 λ
(1 − P0 )
H K
4. The expected number of customers in the system:
M µ
Ls = ∑ nPn = Lq + (1 − P0 ) = M −
n =0 λ
1 − P0 d i
5. The expected waiting time of a customer in the queue:
Lq
Wq =
λ M − Ls b g
6. The expected waiting time of a customer in the system:
1 Ls
Ws = Wq + or
µ λ( M − L ) s

Model VII: {(M/M/s) : (M/GD)} Multiserver – Finite Population (Source) of Arrivals

When the number of servers are more than one (i.e. s > 1), the steady-state equations are derived in the
same way as in Models IV and V.
Mρ P0 = P1 ; n=0
{( M − n)ρ + n} Pn = ( M − n + 1) ρ Pn − 1 + (n + 1) Pn + 1 ; 0≤n≤ M
{( M − n)ρ + s} Pn = ( M − n + 1) ρ Pn −1 + sPn +1 ; s < n ≤ M −1
sPM = ρ Pn −1 ; n=M
We define λ n and µ n as the rate of arrival and service, respectively as follows:

λn =
RS( M − n)λ ;0≤n< M
T0 ;n≥ M

µn =S
Rnµ ;0≤n< s
T sµ ; n≥s
Again substituting for λ n and µ n in the expression for P0 and Pn in Model V(A), we get:
R| M ! F λ I n

| G JP
n!( M − n)! H µ K
;0≤n≤s
0
P =S
n
|| M ! FG λ IJ P ; s < n ≤ M n

T ( M − n)! s! s H µ Kn−s 0

R| s −1 M! FG λ IJ + ∑ n
M! M FG λ IJ n U| −1

P =S∑ V|
0
|T n=0 n ! ( M − n ) ! H K
µ n ! ( M − n) ! s ! s
n=s
n−s
H µK W
Performance Measures of Model VII
1. The expected number of customers in the queue:
M M M
| Lq = ∑ (n − s) Pn = ∑ nPn − s ∑ Pn
n = s +1 n = s +1 n= s+ 1
M
= Σ nPn − Σ nPn − s
s RS Σ P − Σ P UV
M s

n=0 n= 0 T n=0 W n
n=0
n

M
= ∑ nPn − ∑ nPn
s R
− s S1 − ∑ P V
U s

n=0 n=0 T W n=0

n

s
= Ls − s + ∑ ( s − n) Pn = Ls − ( s − s )
n=0

b g
s
1. where s = expected number of idle servers = ∑ s − n Pn .
n=0
594 Operations Research: Theory and Applications

2. The expected number of customers in the system:

λ eff
Ls = Lq + ( s − s ) = Lq + , where λ eff = µ ( s − s )
µ
The expression for λeff represents the expected arrival rate λ (M – n) of customers (n customers are
already in the system each with mean arrival rate λ) under steady-state conditions. Thus
M M M
λ eff = ∑ λ ( M − n) Pn = λ M ∑ nP0 − λ ∑ nPn
n=0 n=0 n=0

= λM − λLs = λ ( M − Ls )
3. The expected waiting time of a customer in the system:
Ls Ls
Ws = =
λ eff λ ( M − Ls )
4. The expected waiting time of customer in the system:

Lq Lq
Wq = =
λ eff λ ( M − Ls )
Example 16.11 A mechanic repairs four machines. The mean time between service requirements is 5 hours
for each machine and forms an exponential distribution. The mean repair time is one hour and also follows
the same distribution pattern. Machine downtime costs Rs 25 per hour and the mechanic costs Rs 55 per day.
Determine the following:
(a) Probability that the service facility will be idle
(b) Probability of various number of machines (0 through 4) to be out of order and being repaired
(c) Expected number of machines waiting to be repaired, and being repaired
(d) Expected downtime cost per day
Would it be economical to engage two mechanics, each repairing only two machines?
[Delhi Univ., MBA, 2003]

Solution From the data of the problem, we have

λ = 1/5 = 0.2 machine/hour, µ = 1 machine/hour,
M = 4 machines, and ρ = λ /µ = 0 . 2
(a) The probability that the system shall be idle (or empty) is:

L
P =M∑
M ! F λI O
n
−1
LM OP −1

MN ( M − n) ! GH µ JK PPQ
M 4 4! n
= ∑ (0.2)
0
n=0 N n=0 ( 4 − n) ! Q
L 4! 4!
= M1 + (0. 2) + (0. 2) 2 4!
+ (0. 2) 3
4!
+ (0. 2) P
O 4
−1

N 3! 2! 1! 0! Q
−1
= 1 + 4 ( 0.2 ) + 4 × 3( 0.04 ) + ( 4 × 3 × 2 ) ( 0.008 ) + ( 4 × 3 × 2 × 1) ( 0.00016 )

−1
= 1 + 0.8 + 0.48 + 0.192 + 0.000384 = (2.481)–1 = 0.4030.

(b) The probability that there shall be various number of machines (0 through 4) in the system [See Table 16.2] is:

Pn =
M! λFG IJ n
P0 ; n ≤ M
b gH K
M −n ! µ
(c) The expected number of machines to be out of order and being repaired
µ 1
| Ls = M − (1 − P0 ) = 4 − (1 − 0.403) = 4 − 2.985 = 1.015 machines
λ 0.2
 Queuing Theory 595

M! λ FG IJ n
n
( M − n )! µ H K Probability*

(1) (2) (3) = (2) × P0

0 1.00 0.4030
1 0.80 0.3224
2 0.48 0.1934
3 0.19 0.0765 Table 16.2
4 0.00 0.0000 Calculation of Pn

* The sum total of these probabilities is 0.9953 instead of 1. It is because of the approximation error.

(d) Expected time a machine will wait in queue to be repaired

1LM M λ+µ OP = L 4 − 0.2 + 1 OP
Wq =
MN
µ 1 − P0
−
λ PQ MN1 − 0.403 0.2 Q
4
= − 6 = 0.70 hours or 42 minutes
0. 597
(e) If there are two mechanics each serving two machines, then M = 2, and therefore,
L FG λ IJ n OP −1

= M ∑
2 M!
P0 = 1 + 2 ( 0.2 ) + 2 × 1( 0 .2 ) 2 = 0.68
MN n=0 ( M − n )! H µK PQ
Cost Analysis : It is assumed that each mechanic with his two machines constitutes a separate system with
no interaction. Thus,
(a) The expected number of machines in the system will be:
µ 1
(
Ls = M − 1 − P0 = 2 −
λ 0.2
)
(1 − 0.68 ) = 0.4 machine
The expected downtime of machines per day is
= Expected number of machines in the system × 8-hour day × Number of mechanics
= 0.4 × 8 × 2 = 6.4 hours/day
Total cost for hiring two mechanics will be:
Total cost = Mechanics’ cost + Downtime cost
= 2 × 55 + 6.4 × 25 = Rs 270 per day
But the total cost with one mechanic is Rs (55 + 200) = Rs 255/day. Hence, it is not economical to engage
two mechanics.
Example 16.12 There are 5 machines, each of which, when running, suffer breakdown at an average rate
of 2 per hour. There are 2 servicemen and only one man can work on one machine at a time. If n machines
are out of order when n > 2 then (n – 2) of them have to wait until a serviceman is free. Once a serviceman
starts work on a machine the time to complete the repair has an exponential distribution with mean of 5 minutes.
Find the distribution of the number of machines out of action at a given time. Also the find average time an
out-of-action machine has to spend waiting for the repairs to start. [AMIE, 2005]

Solution From the data of the problem, we have

λ = 2 per hour, µ = 60/5 per hour, M = 5 machines, and s = 2 servicemen
−1
2 −1 5! 2
n 5 5!  2 
n
648
Then P0 =  ∑   + ∑    =
 n = 0 (5 − n) ! n !  12  n = 2 (5 − n ) ! 2 ! 2n − 2  12   1, 493

R| 5 ! F 2 I F 648 I P n

P =S
| (5 − n) ! n ! GH 12 JK GH 1493JK 0 ; 0≤n<2
and n
|| 5! FG 2 IJ FG 648 IJ P
n

T (5 − n ) ! 2 ! 2 n−2 H 12 K H 1493K 0 ; 2≤n≤5

596 Operations Research: Theory and Applications

(i) Average number of machines out-of-action at a given time

5 165
Lq = ∑ ( n − 2 ) Pn = P3 + 2 P4 + 3P5 = = 0.110 hour.
n = 2 +1 1,493
(ii) Average waiting time an out-of-action machine in the workshop
Lq
Wq =
λ eff
5
where λ eff = λ ∑ ( 5 − n ) Pn = λ ( 5 P0 + 4 P1 + 3 P2 + 2 P3 + P4 )
n=0
648
= 2(5P0 + 3.333P0 + 0.833P0 + 0.023P0) = (2) (9.189) = 8 machines (approx.)
1,493
0110
.
Therefore | Wq = = 0.013 hour
7.976

SELF PRACTICE PROBLEMS C

1. A group of engineers has two terminals to aid in their calculations. (c) Would it be desirable to provide two mechanics, each to
The average computing job requires 20 minutes of terminal time, service only two machines?
and each engineer requires some computation, about once every 3. In a small handloom mill there are four looms working continu-
0.5 hour, i.e. the mean time between calls for service is 0.5 hours. ously. Occasionally, the looms breakdown for which a repairman
Assume these are distributed according to an exponential distri- is called. On an average the service requirement is at 10-hour
bution. If there are six engineers in the group, find: intervals which follows an exponential distribution. The average
(a) the expected number of engineers waiting to use one of the repair time of the looms is Rs 20 per hour and the cost of the
terminals mechanic is Rs 50 per day. What is the cost of the present
(b) the total lost time per day system? Is it desirable to have two mechanics? Should we
[Hint: λ = 2 , µ = 3, m = 6 and s = 2 ; P0 = 0.0268; allocate two looms to each mechanic or all the four looms should
be allocated to the two mechanics together?
(i) Lq = 1.40, (ii) Time lost per day = 8(1.40) = 11.2 hours]
4. At a port, there are six unloading berths and four unloading crews.
2. A mechanic services four machines. For each machine, the mean When all the berths are full, the arriving ships are diverted to an
time between service requirements is 10 hours and is assumed overflow facility 20 miles down the river. Tankers arrive according
to form an exponential distribution. The repair time tends to follow to a Poisson process with a mean of one every 2 hours. It takes
the same distribution with a mean of two hours. When a machine the unloading crew, on the average, ten hours to unload a tanker,
is down for repairs, the time lost has a value of Rs 20 per hour. the unloading time following an exponential distribution. Find:
The mechanic costs Rs 50 per day. Given this information find: (a) On an average, how many tankers are at the port?
(a) What is the expected number of machines in operation? (b) On an average, how long does a tanker spend at the port?
(b) What is the expected downtime cost per day? (c) What is the average arrival rate at the overflow facility?

16.9 MULTI-PHASE SERVICE QUEUING MODEL

Model VIII: {(M/Ek / 1) : ( ∞ / FCFS)} Erlang Service Time Distribution
with k-Phases
This model consists of single service channel in which there are k identical stages (phases) in series for
services, each with average service time 1/kµ, as shown in Fig. 16.9. The distribution of total servicing time
of a customer in the system is the combined distribution of time in all these phases.
Since each customer is served in k-phases one-by-one and a new service does not start until all k-phases
have been completed, therefore, each arrival increases the number of phases by k in the system. Thus, if there
are m customers waiting in the queue and one customer is already in service at sth phase, then the total number
of phases in the system (waiting and in service) will be n = mk + s.

Fig. 16.9
Service Channels
in Series
 Queuing Theory 597

If µ denotes the number of customers served per unit of time, then kµ will be the number of phases
served per unit of time, and 1/kµ will be the average time taken by server at each phase. Therefore:
λ n = λ phases arrive per unit time
µ n = kµ phases served per unit time
The probability density function for Erlang distribution is:
( kµ ) k k − 1 − k µt
f (t ) = t e ; t ≥ 0
( k − 1) !
where µ = expected number of customers completing service per unit of time Erlang distribution
is a service time
k = a positive integer distribution whose
The expected total service time and variance of Erlang distribution is given by: shape parameter k
specifies the amount

E (T ) = k
FG 1 IJ = 1 σ2 = k
FG 1 IJ 2
1
variability in the
service times.
H kµ K µ and
H kµ K =
kµ 2
For general value of k, the model value of service time, t is (k – 1)/kµ.
Figure 16.10 shows how the shape of the Erlang distribution changes for various values of k, when
k = 1, it reduces to the exponential distribution, whereas for 1 < k < ∞ , it reduces to a constant distribution
for customer inter-arrival times.

Fig. 16.10
Erlang
Distribution

The system of steady-state difference equations for this distribution is obtained in the same way as
discussed earlier. [See Appendix 16.A for proof ]
λ P0 = kµ P1 ; n=0
( λ + kµ) Pn = λPn − k + k µPn + 1 ; n ≥1
where, P0 = 1 − ρk .

Performance Measures of Model VIII

1. The expected number of phases (not customers) in the system

L (k ) k +1 λ
Lq ( k ) = s =
µ 2 µ(µ − λ )
k +1 λ
Ls ( k ) =
2 µ−λ
2. The expected number of customers (not phases) in the queue
Ls ( k ) − Average number of phases in service
Lq =
k
LM
1 k +1 λ k +1 λ OP = FG k + 1 IJ F λ I
2

Q H 2 k K GH µ ( µ − λ ) JK
= −
k N 2 µ−λ 2 µ
Since 1/µ is the average service time per customer and (k + 1)/2 is the average number of phases of
one customer in service, therefore the time taken for serving a customer will be (k + 1)/2µ. Thus, the average
number of phases that arrive in this time would be λ ( k + 1)/2 µ .
598 Operations Research: Theory and Applications

3. The expected waiting time of a customer in the queue

Lq k +1 λ
Wq = =
λ 2 k µ (µ − λ )
4. The expected waiting time of a customer in the system
1 k +1 λ 1
Ws = Wq + = +
µ 2 k µ (µ − λ ) µ
5. The expected number of customers in the system
λ k + 1 λ2 λ
Ls = Lq + = + or Ls = λWs
µ 2 k µ (µ − λ ) µ
Example 16.13 In a factory cafeteria the customers (employees) have to pass through three counters. The
customers buy coupons at the first counter, select and collect the snacks at the second counter and collect
tea at the third. The server at each counter takes, on an average, 1.5 minutes, although the distribution of
service time is approximately Poisson at an average rate of 6 per hour. Calculate
(a) The average time a customer spends waiting in the cafeteria.
(b) The average time of getting the service.
(c) The most probable time in getting the service.
Solution From the data of the problem, we have
λ = 6 customers/hour; Service time per phase = 15 minutes
µ = 4.5 (= 1.5 × 3) customer/minute or 13.34/hour; k = 3
(a) The average time a customer spends waiting in the cafeteria
k +1 λ 3 +1 6 9
Wq = = = hour or 2.43 minutes
2k µ(µ − λ ) 2(3) 13.34(13.34 − 6) 220
(b) The average time of getting the service is the average of the time when it is following the third phase of
service. Thus, the average time spent in getting the service is:
1 1
= hours or 4.50 minutes
µ 13.34
(c) The most probable time spent in getting the service is the model value of service time for the third phase
of service. Thus, most probable time spent is
k −1 3−1 1
= = hour or 3 minutes
kµ 3 × 13.34 20
Example 16.14 An airline maintenance base has facilities for overhauling only one aeroplane engine at
a time. Hence, to return the aeroplanes into use at the earliest, the policy is to stagger the overhauling of the
4 engines of each aeroplane. In other words, only one engine is overhauled each time an aeroplane comes
into the base. Under this policy, aeroplanes have arrivals according to a Poisson process, at a mean rate of
one per day. The time required for an engine overhaul has an exponential distribution, with mean of half day.
A proposal has been made to change the policy so as to overhaul all four engines consecutively each
time an aeroplane comes into the shop. It is pointed out that although this will quadruple the expected service
time, each plane would need to come into the shop only one-fourth time as often. Compare the two alternatives
on a meaningful basis.
Solution The two alternatives will be compared on the basis of the cost of waiting time cost of the
aeroplanes that require overhauling.
First alternative : {(M/M/1) : ( ∞ /FCFS)} queuing system
Given that λ = 1 aeroplane per day; µ = 2 aeroplanes per day
Therefore, the average number of aeroplanes in the system is
λ 1
Ls = = =1
µ− λ 2 −1
Second alternative : {(M/Ek/1) : (∞/FCFS)} queuing system
Given that λ = 1/4 aeroplane per day; k = 4
Since service time per aeroplane is 4 × (1/2) = 2 days, therefore, the mean service rate, µ = 1/2 aeroplane per
day. Thus average number of aeroplanes in the system are
 Queuing Theory 599

k +1 λ2 λ 4 +1 (1/4 ) 2 1/4 13
Ls = + = ⋅ + = or 0.81
2k µ(µ − λ) µ 2(4) 1/2(1/2 − 1/4 ) 1/2 16
Since Ls (= 0.81) in the second alternative is less than its value in the first alternative, therefore the waiting
cost for requiring overhauling in the second alternative will be less. Hence, the proposal should be accepted.
Example 16.15 At a certain airport it takes exactly 5 minutes to land an aeroplane, once it is given the
signal to land. Although incoming planes have scheduled arrival times the wide variability in arrival times
produces an effect which makes the incoming planes appear to arrive in a Poisson fashion at an average rate
of 6 per hour. This produces occasional stockups at the airport that can be dangerous as well as costly. Under
these circumstances, how much time will a pilot expect to spend circling the field waiting to land?
Solution From the data of the problem, we have
λ = 6 per hour or 1/10 per minute; µ = 1/5 per minute, and k = ∞, as service time is constant
Hence, the average time that a pilot expects to spend circling the field, waiting to land, is given by
k +1 λ 1 1 FGλ IJ
Wq = lim
k →∞
= lim
2 k µ (µ − λ ) k → ∞ 2
1+
H
k µ (µ − λ ) K
1FG 1 IJ 1/10
=
2 H
1+
∞ K 1/5( 1/5 − 110
/ )
= 5/2 or 2.5 minutes.

SELF PRACTICE PROBLEMS D

1. A hospital clinic has a doctor examining every patient brought load, get to the customer, unload and return to the warehouse.
in for a general check-up. On an average, the doctor spends Loading and unloading times are small as compared to the travel
4 minutes on each phase of the check-up, although the distri- time.
bution of time spent on each phase is approximately exponential. 4. Repairing a certain type of machine that breaks down in a given
If each patient goes through four phases in the check-up and factory consists of five basic steps that must be performed
if the arrivals of the patients at the doctor’s office are approxi- sequentially. The time taken to perform each of the given steps
mately Poisson at the average rate of three per hour, what is is found to have an exponential distribution with mean 5 minutes
the average time spent by a patient waiting in the doctor’s office? and is independent of the other steps. If these machines break
What is the average time spent in the examination? What is the down in a Poisson fashion at an average rate of two per hour,
most probable time spent in the examination? and if there is only one repairman, what is the average idle time
2. A barber with a one man shop takes exactly 25 minutes to for each machine that has broken down?
complete one hair cut. If customers arrive in a Poisson fashion 5. In a car manufacturing plant, a loading crane takes exactly 10
at an average rate of one every 40 minutes, how long on an minutes to load a car into a wagon and again comes back to
average must a customer wait for service. the position to load another car. If the arrival of cars is in a
[ Meerut, MSc (Maths), 2002] Poisson stream at an average rate is one after every 20
3. A warehouse in a small state receives orders for a certain item minutes, calculate the average waiting time of a car in the queue.
and sends them by a truck as soon as possible to the customer. 6. A colliery working one shift per day uses a large number of
The orders arrive in a Poisson fashion at a mean rate of 0.9 per locomotives that breakdown at random intervals i on an average
day. Only one item at a time can be shipped by truck from the one fails per 8-hour shift. The fitter carries out a standard
warehouse that is located in central part of the state. The maintenance schedule on each faulty locomotive. Each of the
distribution of service time in days has distribution with the five main parts of this schedule takes, on an average, half-an-
probability density function 4e–2t. What is the expected delay hour but the time varies widely. How much time will the fitter
between the arrival of an order and the arrival of the item to the have for the other tasks and what is the average time a
customer? Service time here implies the time the truck takes to locomotive is out of service?

HINTS AND ANSWERS

1. λ = 1/20; µ = 1/16, and k = 4 k +1 λ 1 1  λ

k +1 λ Wq = Lim = Lim  + 
(i) Wq = = 40 minutes k →∞ 2 k µ (µ − λ ) k →∞  2 2 k  µ (µ − λ )
2k µ(µ − λ )
= 20.8 minutes; 1/∞ = 0
(ii) Average time spent in the examination, 1/µ = 16 minutes
(iii) Most probable time spent in the examination (mode value of Ws = Wq + 1 / µ = 45.8 minutes
k −1 3. Erlang probability density function in this case is given by
t for the fourth member of the Erlang family = = 12
kµ ( kµ ) k k −1 − kµt
minutes. f (t , µ, k ) = t ⋅e = 4 te − 2 t
( k − 1) !
2. λ = 1/40 , µ = 1/25, and k → ∞ (because service is constant)
600 Operations Research: Theory and Applications

This gives µ = 1, and k = 2. This means the service time 5. λ = 3/hour, µ = 6/hour, k → ∞ (because service is constant)
distribution is a second member of the Erlang family with µ = 1.
Also, given that λ = 0.9 customer per day. Thus
k +1 λ 1FG 1IJ λ

FG 1 + 1 IJ λ = 6.75 days
Wq = lim
k→∞ 2k µ(µ − λ)
= lim
k→∞ 2 H
1+
K
k µ(µ − λ)
Wq =
H 2 2k K µ(µ − λ) = 5 minutes, (1/ ∞ = 0)
1 6. λ = 1/8 per hour, since a fitter takes on an average 5 × (1/2) = 5/
Ws = Wq + = 7 .75 days 2 hours to repair a locomotive, therefore µ = 1/(5/2) = 2/5 and the
µ
time the fitter takes to complete the task is 8 – (5/7) = 5.5 hours.
4. λ = 2/60, µ = 1/(5 × 5) = 1/25, k = 5, and Ws = Average time a locomotive is out of service
λ /kµ = 5/(5 × 30) = 0.166
k +1 λ 1
Expected idle time = Average time spent in the system, = ⋅ + = 2 .18 hours
2k µ(µ − λ) µ
1 k +1 λ 1
Ws = Wq + = + = 100 minutes.
µ 2k µ(µ − λ ) µ

16.10 SPECIAL PURPOSE QUEUING MODELS

Model IX: Single Server, Non-Exponential Service Times
Distribution – Unlimited Queue

When service time cannot be described by an exponential distribution, the normal distribution could also
be used to represent the service pattern of a single server queuing system. A queuing model where arrivals
form a Poisson process, while the service times follow normal distribution depends on the standard
deviation for service time and assumes no particular form for the distribution itself. The performance
measures in this case are determined as under:
λ
P0 = 1 –
µ
λ2 σ 2 + ( λ /µ ) 2 λ
Lq = 2 (1 − λ /µ ) ; Ls = Lq +
µ
Lq 1
Wq = ; Ws = Wq +
λ µ

It may be noted that the values of these performance measures, except P0, depend on standard
deviation (σ) of the service time distribution. Since σ appears in numerator, therefore, greater variability in
the service time will result in longer waiting time. Thus, consistency in service time is very important for
ensuring overall quality of the service period.
Example 16.16 At one man barber shop, customers arrive at a mean rate of 4 per hour. The customers
are served at a mean rate of 5 per hour. The owner feels that service times have some unspecified positive
skewed unimodal two-tailed distribution with a standard deviation of σ = 0.05 hour (3 minutes).
(a) Determine the queuing characteristics for barber shop.
(b) How much would the assumption of exponential service times distort these values. Discuss.
Solution (a) From the data of the problem, we have λ = 4, µ = 5 and σ = 0.05.
If owner uses the single sever model with exponential arrival and non-exponential service times, then the
values of measures of performance for the model are:

Lq =
b
λ2 σ 2 + λ / µ g 2
=
a 4 f a 0 . 05f + b 4/5g
2 2 2
= 1. 7
b
2 1− λ /µ g 2 b1 − 4 /5g
2

Ls = Lq + λ/µ = 1.7 + 4/5 = 2.5

Wq = Lq /λ = 1.7/4 = 0.425
Ws = Wq + 1/µ = 0.425 + 1/5 = 0.625
(b) If service times follow exponential distribution, then the results obtained are: |Lq = 3.2, Ls = 4,
Wq = 0.8 and Ws = 1. It may be noted here that the customer’s waiting time and the number of customers are
more than those in previous case.
 Queuing Theory 601

Model X: Single Server, Constant Service Times – Unlimited Queue

If the service time is constant (= 1/µ ) instead of exponential distribution time, for serving each customer,
then the variance σ2 = 0 and obviously, the values of Ls, Lq, Ws and Wq will be less than those values in
the models discussed before.
Substituting σ 2 = 0 in Model IX, we get
(λ /µ ) 2 λ2 λ
Lq = = ; Ls= Lq +
2{1 − (λ /µ )} 2µ ( µ − λ ) µ

Lq λ 1
=
Wq =
λ b
2µ µ − λ g ; Ws= Wq +
µ
The assumption of constant service time puts an absolute lower bound on the value of the mean queue
length and the upper bound can be obtained by assuming exponential service because the negative
exponential distribution is the high-variance distribution.
Example 16.17 Truck drivers who arrive to unload plastic materials for recycling currently have to wait
an average of 15 minutes before unloading their trucks. The cost of driver and truck time wasted while in queue
is valued at Rs 200 per hour. A new device is installed to process truck loads at a constant rate of 10 trucks
per hour at a cost of Rs 10 per truck unloaded. Trucks arrive according to a Poisson distribution at an average
rate of 8 per hour. Suggest whether a device should be put to use or not. [AMIE, 2005]

Solution Waiting cost before the use of new device

Waiting cost = Waiting time × Cost of waiting = (15/60) × 200 = Rs 50 per trip
Under new system λ = 8 per hour, and µ = 10 per hour
λ2 8
Wq = = = 1/5 hour = 12 minutes
b
2µ µ − λ g b gb
2 10 10 − 8 g
Waiting cost per trip with new device = (1/5) × 200 = Rs 40 per trip
Saving with new device = Current system – New system = 50 – 40 = Rs 10 per trip
New savings = Savings – Cost of installing new device = (10 – 5) = Rs 5 per trip.

Model XI: Service Level Cost Model

As shown in Fig. 16.1, service level is the function of two conflicting costs:
(i) Cost of offering the service to the customers
(ii) Cost of delay in offering service to the customers
This means the customer’s service level determines the efficiency of the queuing system. Hence, an optimum
service level that must be maintained by a queuing system can be determined by determining the optimum
level of service rate (µ). This rate should offer service and number of service facilities (s, servers) in a way
that avoids excessive delay in offering the service.
In this model, we will discuss the procedure of determining the optimum values of both service rate (µ)
and number of service.
Optimum service rate The procedure discussed here for determining optimum service rate (µ) is based on
Model I. Let
C1 = cost per server per unit of time, i.e. cost per unit of time having a server available
C3 = cost of waiting per customer per unit time, i.e. cost per unit of time having a customer wait
in the system
TC(µ) = total expected cost of waiting and service per unit time, given the service rate µ.
Then, the cost equation can be written as
TC(µ) = (Service rate × Cost per service per unit time) + (Number of customers in the system ×
((Waiting cost per customer per unit time)
λ λ
= µC1 + Ls C2 = µC1 + C2 ; Ls = (Model I)
µ−λ µ−λ
602 Operations Research: Theory and Applications

Since the service rate is measured in continuous units of time, its optimum value can be obtained by using
the concept of maxima and minima in differential calculus. Thus, we get:
d µC2
TC(µ ) = C1 − = 0
dµ (µ − λ ) 2
as the necessary condition for maximum or minimum value of µ. Thus, we further get:
µ=λ+ λC2 /C1
This value of µ represents its optimum value because the second derivative of TC(µ), with respect to µ, is
positive for µ > λ.

Remark If a queuing system has limited capacity of N customers (see Model III), then TC(µ) can be modified
as:
TC( N , µ) = µC1 + Ls ⋅ C2 + N . C3 + λPN ⋅ C4
where C3 = cost of serving each additional customer per unit of time
C4 = cost per lost (balking) customer (i.e. λPN), i.e. cost of losing a customer.
Optimum number of service facilities (servers) For determining the optimum number of service
facilities (servers) we will consider the operational characteristics of Model IV(A): {(M/M/s) : ( ∞ /FCFS)}.
In this case the cost equation can be written as:

TC(s) = sC1 + Ls(s) . C2

where Cs = cost per server per unit of time
Cw = cost of waiting per customer per unit of time
Ls(s) = expected number of customers with system when there are s number of servers.
Since s can be measured only in discrete units, therefore, its optimum value can be obtained by using the
concept from difference equations. That is, the necessary condition for a minimum of the given function is:
∆TC( s − 1) < 0 < ∆TC( s)
The condition ∆TC ( s − 1) < 0, gives
TC( s) − TC( s − 1) < 0,
or s. Cs + Ls ( s). Cw − {( s − 1)Cs + Ls ( s − 1). Cw} < 0
Cs
L s ( s − 1) − L s ( s ) >
Cw
The condition ∆TC( s) > 0 gives
TC( s − 1) − TC( s) > 0
{( s + 1)Cs + Ls ( s + 1)Cs } − {sCs + Ls (s )Cw } > 0
These two results, together, yield the condition to determine optimum value of s.
Cs
Ls − Ls ( s + 1) ≤ ≤ Ls ( s − 1) − Ls ( s)
Cw
Example 16.18 The average rate of arrivals at a service store is 30 customers per hour. At present there
is one cashier who, on an average, attends 45 customers per hour. The store owner estimates that each extra
minute of system process time per customer means an additional cost of Rs 1.5. An assistant can be provided
to the cashier and in that case the service unit can deal with 85 customers per hour. Determine the rate
(customers per hour) with which customers should be attended in order to minimize customer waiting cost
and to justify the employment of an assistant.
Solution Given that, λ = 30 customers/hour, µ = 45 customers/hour and cost per additional customer
service, C1 = Rs 1.5. The cost of waiting per hour is as follows:
Cost of waiting = Expected number of customers in the system × Hourly waiting cost
λ 30
= × 15
. × 60 = × 1. 5 × 60 = Rs 180/hour.
µ−λ 45 − 30
 Queuing Theory 603

Therefore, C2 = Rs 180/hour.
The new value of service rate is obtained as:
λC2 30 × 180
µ=λ + = 30 + = 90 customers/hour.
C1 1.5

CHAPTER SUMMARY

Waiting line problems are commonly found in production and service systems which often face random arrival rates and service times.
The management of these may change the quality of life and the productivity of various systems.
Key components of a queuing system are the arriving customers, the queue in which they wait for service, and the servers that
provide the service. A queueing system needs to specify the number of servers, the distribution of interarrival times, and the
distribution of service times. An exponential probability distribution is chosen for the distribution of interarrival times because of
random arrival of customers at the queuing system. Other probability distributions used for the service-time distribution include
constant service times distribution and the Erlang distribution.
The measures of performance of queuing systems are the expected values of the number of customers in the queue or in the system
and of the waiting time of a customer in the queue or in the system. In addition to the expected values, the probability distributions
of these quantities are used as measures of performance.

CHAPTER CONCEPTS QUIZ

True or False 19. The payment of telephone bills by cheque or cash is the
example of _______ discipline.
1. Customer population is one of the characteristic of any queuing 20. The service facilities, commonly known as _______ may be in
system. series or in parallel.
2. Utilization factor is the key operating characteristic for a queuing
system. Multiple Choice
3. First-come-first-service is the priority queue discipline.
21. Customer behaviour in which the customer moves from one
4. When capacity of any queuing system is infinite, the calling
population is called infinite. queue to another in a multiple channel situation is
(a) balking (b) reneging
5. The average waiting time of customers in the system is not used
for economic analysis of a queuing system. (c) jockeying (d) alternating
6. A calling population is considered to be infinite when arrivals of 22. Which of the following characteristics apply to queuing system
customers is not restricted. (a) customer population (b) arrival process
7. As arrival rate at any queuing system increases, the cost of (c) both (a) & (b) (d) neither (a) nor (b)
providing service decreases. 23. Which of the following is not a key operating characteristic for a
8. Total number of customers in the queuing system is one of the queuing system
components of service mechanism. (a) utilization factor
9. A customer who does not switch between waiting lines is called (b) per cent idle time
a patient customer. (c) average time spent waiting in the system and queue
10. The dynamic arrival process in controlled by both the service (d) none of the above
facility and the customers.
24. Priority queue discipline may be classified as
(a) finite or infinite (b) limited and unlimited
Fill in the Blanks (c) pre-emptive or non-pre-emptive (d) all of the above
11. Customers that require service are generated it different times 25. Which symbol describes the interarrival time distribution
by a _______, commonly known as input source. (a) D (b) M
12. Customers, from a queue, are selected for service according to (c) G (d) all of the above
certain rules known as _______.
26. Which of the following relationships is not true
13. In the _______, the arrival of customers depends on the nature (a) Ws =LWq + 1/µ (b) Ls = λWs
of arrival rate.
(c)WLs =WLq + 1/λ (d) Lq = λWq
14. The _______ is controlled by both the service facility and the
customers. 27. The calling population is assumed to be infinite when
(a) arrivals are independent of each other
15. The _______ distribution provides probabilities for times gap
between two consecutive arrivals. (b) capacity of the system is infinite
16. The interarrival time is _______ by an exponential distribution, (c) service rate is faster than arrival rate
also called _______. (d) all of the above
17. The mean of exponential distribution is the _______ time be- 28. Which of the cost estimates and performance measures are not
tween arrivals. used for economic analysis of a queuing system
18. The _______ refers to the number of queues and their respec- (a) cost per server per unit of time
tive length. (b) cost per unit of time for a customer waiting in the system
604 Operations Research: Theory and Applications

(c) average number of customers in the system 32. Expected length of non-empty queue is given by
(d) average waiting time of customers in the system (a) L = µ / (µ – λ) (b) L = sµ / (sµ – λ)
29. A calling population is considered to be infinite when (c) L = λ /(µ – λ) (d) λ/(µ – λ) + (1/µ)
(a) all customers arrive at once 33. Cumulative probability distribution of waiting time for customer is
(b) arrivals are independent of each other given by
(c) arrivals are dependent upon each other (a) ρe–µt (1–ρ) (b) (1 – ρ)e – µ t (1–ρ)
– λ t (1– ρ)
(d) all of the above (c) (1 – ρ)e (d) (1 – ρ)e – λ t (1+ ρ)
30. The cost of providing service in a queuing system decreases 34. The potential loss of customers is given by
with s s
(a) decreased average waiting time in the queue
1 λ  1 λ
(a) Ps =   P0 (b) Ps =   P0
(b) decreased arrival rate s !  sµ  n!  µ 
(c) increased arrival rate s
1 λ
(d) none of the above (c) Ps =   P0 (d) none of the above
s! µ 
31. Service mechanism in a queuing system is characterized by
(a) server’s behaviour (b) customer’s behaviour 35. Expected waiting time of customer in the system is
(c) customers in the system (d) all of the above (a) Wq = Lq /λeff (b) Wq = Ls /λeff
(c) Wq = Ls – (1/µ) (d) Wq = Lq + (λeff /µ)

Answers to Quiz
1. T 2. F 3. F 4. F 5. T 6. T 7. F 8. F 9. F 10. T
11. calling population 12. queue discipline 13. static arrival pattern 14. dynamic arrival process
15. exponential 16. approximated, negative exponential distribution 17. expected or mean,
18. queuing process 19. dynamic queue, 20. service channels.
21. (c) 22. (c) 23. (d) 24. (c) 25. (d) 26. (c) 27. (a) 28. (d) 29. (b) 30. (d)
31. (a) 32. (a) 33. (b) 34. (c) 35. (a)

CASE STUDY

Case 16.1: Broadcasting Corporation

Broadcasting Corporation (BC) is a service-oriented organization that has a prime objective of reaching more and more
people in the country through the medium of radio. Today, it has broadcasting centres at 80 places and is providing
a coverage of 90 per cent of population in the country. In addition to this, external broadcast in 20 languages are also
provided. Being the only broadcasting organization, the demand for expansion of its services is very high. Broadcast
signals are not received in some areas and because of use of very high powered transmitters by the neighbouring
countries, signals at some of the existing places are also not very satisfactory.
The government has now decided to expand the network of BC and Rs 700 crore are allocated for the provision
of about 150 new transmitting stations.
Keeping in view the latest state-of-art in broadcasting and also considering the fact that the equipment installed
now will continue to service for another 20 to 25 years, it has been decided to instal all latest technical equipment
at these centres.
The requirement of technical personnel for the operation and maintenance of these equipment is very large. Trained
personnel in this field are not available and time is not available for the training of fresh recruits for the repair and
maintenance of this sophisticated equipment.
It has, therefore, been decided that instead of following the present policy of maintenance at every station,
centralized maintenance policy will be ideal. This is even otherwise most essential considering the high cost of test
and measuring equipment required for the repair/maintenance of this equipment. Only operational staff will, therefore,
be provided at all these stations. At the maintenance centres all automatic test and measuring equipment and highly
trained technical staff will be provided so that the new equipment can be maintained most effectively and efficiently.
Each maintenance centre will cater to the needs of about 4 to 5 adjacent stations so that there is no delay in the
provision of service for these centres.
The problem is to work out the minimum possible and sufficient requirement of technical staff for each
maintenance centre so that the service can be provided most efficiently and economically. Though sufficient spare units
will be kept at each station, however, the defective units will be required to be repaired immediately so that
uninterrupted service can be extended to the public. Some of the equipment are very vital and cannot wait long for
repair/maintenance, while a few others are not that crucial and can wait for sometime for repair. But once the equipment
is opened for repair, it should be completed before taking up the new equipment of most priority so that there is
no loss of valuable components etc.
 Queuing Theory 605

Currently, this problem is being faced for the first time and instead of the usual method of ad hoc provision for
staff, it will be preferred if the requirements of staff for repair/maintenance are decided to be based on scientific method
so that there is neither an excess of staff nor the equipment has to wait unnecessarily long for want of servicing
personnel.
The data for the number of units required to be repaired, their relative priority, likely rate of failure, waiting time,
etc., are shown in the following pages:

Name of No. of No. of Total Number of Likely Rate Total Failure Rate
Equipment Units per Modules Modules per of per Maintenance
Station per Unit Maintenance Failure Centre
Centre

Priority 1: Maximum waiting time including service < 0.75/day –

(i) Tape decks 2 5 10 0.1% unit/day/station
(ii) Turn tables 2 5 10 0.5 unit/day
20 × 5 = 100
Priority 2: Maximum waiting time including service < 1 day –
(i) Tape-recorders 3 10 30
(ii) Mixers 2 10 20
(iii) Amplifiers 5 2 10 1.6 units/day/station 8 units/day
(iv) Cass. Decks 5 1 5
(v) UPTS 5 3 15
80 × 5 = 400
Priority 3: Maximum waiting time including service < 1 day
(i) Intercom 5 1 5 0.3 unit/day/station 1.5 units/day
(ii) Misc 1.5 – 5
10 × 5 = 50

Service rate (average) = 2 units/day. Based on the data given above, suggest a queuing system to determine the number
of service personnel that should be recruited by the Corporation.

Case 16.2: Warehousing Corporation

A single crew is provided for unloading and/or loading each truck that arrives at the loading deck of a warehouse. These
trucks arrive according to a Poisson input process at a mean rate of one per hour. The time required by a crew to
unload and/or load a truck has an exponential distribution (regardless of the crew size). The expected time required
by a one man crew would be two hours.
The cost of providing each additional member of the crew is Rs 10 per hour. The cost that is attributable to
having a truck not in use (i.e. a truck standing at the loading deck) is estimated to be Rs 40 per hour. Assume that
the mean service rate of the crew is proportional to its size, what should the size be in order to minimize the expected
total cost per hour?
The general routine is that order pickers assemble orders placed by stores and load them on one of the company
trucks waiting at the single loads, on a first-come, first-served basis. They then proceed to the store for which the
order is destined, unload and return for another order. Because of the many different routes and distances, and traffic
problems at different times of the day, the time between arrivals of the trucks at the dock is random (assumed to
follow an exponential distribution) averaging 30 minutes. The loading time also follows an exponential distribution and
averages 15 minutes.
Truck drivers are paid Rs 10 per hour and the crew of two loaders are each paid Rs 4 per hour. Truckers have
complained about the long waiting, so a sample was taken showing that truckers did indeed wait for an average of
32 minutes. The warehouse manager knows that a second truck dock would probably solve the problem, but the large
capital expenditure plus the disruption of operations during construction are deterrents to this solution.
Tests are made with different crew patterns and it is found that a crew of three can be used to advantage, reducing
the loading time to 10 minutes.
As management trainee working with the corporation, help the corporation in finding suitable answers to the
following questions:
(a) Is the crew of three loaders more economical than the crew of two?
(b) What is the probability that an arriving truck will find at least one truck already in the system?
606 Operations Research: Theory and Applications

(c) How much of the time is the crew idle?

(d) If another truck dock is available with the present operating characteristics, how will it affect the total expected
cost per hour?

APPENDIX 16.A: PROBABILITY DISTRIBUTION OF ARRIVALS

AND DEPARTURES
Theorem 16.1 If the arrivals are completely random, then the probability distribution of number of arrivals in a
fixed time interval follows a Poisson distribution.
Proof Let us define the terms that are commonly used in the development of various queuing models:
∆ t = a time interval so small that the probability of more than one customer’s arrival is negligible,
i.e. during any given small interval of time ∆t only one customer can arrive.
λ ∆t = probability that a customer will arrive in the system during time ∆t.
1 – λ ∆t = probability that no customers will arrive in the system during time ∆t.
Case I : n ≥ 1, and t ≥ 0
Sometimes customers arrive and join the queue even before the start of the service. They may have presumed that their
early arrival would reduce their waiting time. Now the objective is to know the number of customers waiting in line
when the service begins. Thus Pn (t + ∆t), the probability of n customers in the system at time t + ∆t, can be expressed
as the sum of the joint probabilities of the following two mutually exclusive and collectively exhaustive cases:
(i) the system contains n customers at time t and there is no arrival during time interval ∆t.
(ii) the system contains (n – 1) customers at time t and there is one arrival and no departure during time interval
∆t.
That is, Pn (t + ∆t) = {(Probability of n customers in the system at time t) × (Probability of no arrival during
time ∆t)} + {(Probability of n – 1 customers in the system at time t) × (Probability
of one arrival during time ∆t )}
= Pn (t) {1 – λ ∆t} + Pn – 1 (t) {λ ∆t} + 0 (∆t) (1)

Case II : n = 0 and t ≥ 0
If there is no customer in the system at time t + ∆t, then there will be no arrival during ∆t. Thus the probability of
no customer in the system at time t + ∆t is given by
P0 (t + ∆t) = (Probability of no customer at time t) × (Probability of no arrival during time ∆t)
= P0 (t) {1 – λ ∆t} (2)
Equations (1) and (2) may be written respectively as:
Pn ( t + ∆ t ) − Pn ( t ) 0( ∆ t )
= – λ Pn (t) + λ Pn – 1 (t) + ; n ≥ 1
∆t ∆t
P0 ( t + ∆ t ) − P0 ( t ) 0( ∆ t )
= – λ P0 (t) + ; n = 0
∆t ∆t
Letting ∆ t → 0 and taking limit on both sides, we have the following system of differential-difference equations
d
Pn′ ( t ) = {P (t)} = – λ Pn (t) + λ Pn – 1 (t) ; n ≥ 1 (3)
dt n
d
P0′ ( t ) = {P (t)} = – λ P0 (t) ; n = 0 (4)
dt 0
Solution of Differential-Difference Equations Equation (4) can be rewritten as:
p 0′ ( t )
= –λ
p0 (t )
Integrating both sides with respect to t, we get
log P0 (t) = – λ t + A (5)
where A is the constant of integration and its value can be determined by using the following initial conditions:

RS1 ;n= 0 ,t = 0
Pn (0) =
T0 ;n≥0 ,t = 0
Substituting t = 0 in Eq. (5), we get P0 (0) = 1. Thus the value of A = 0. Now Eq. (5) reduces to the form
log P0 (t) = – λ t or P0 (t) = e– λ t ; t ≥ 0 (6)
 Queuing Theory 607

Now, putting n = 1 in Eq. (3) and by using Eq. (6), we get

P1′ ( t ) = – λ P1 (t) + λ P0 (t) (7)
or P1′ ( t ) + λ P1 (t) = λ e– λ t (8)
Equation (8) is a linear differential equation of the first order. Therefore, it can be solved by multiplying both
sides of it by integrating factor:

Integrating factor (IF) = exp e z λdt j = exp b λt g

Then Eq. (8) becomes

m r
e λ t P1′( t ) + λ P1 ( t ) = λ or
d
dt
n
e λ t P1 ( t ) = λ s
On integrating both sides with respect to t, we get
eλ tP1 (t) = λ t + B (9)
where B is the constant of integration and its value can again be obtained by initial conditions. That is, setting t =
0 in Eq. (9), we get P1(0) = B = 0 ; since P1(0) = 0. Thus Eq. (9) reduces to the form
λt
eλ tP1 (t) = λ t or P1 (t) = = λ t e– λ t (10)
eλt
Again, putting n = 2 in Eqn. (3) and using the result of Eq. (10), we get
P2′ ( t ) + λP2 (t) = λ (λ t e– λ t )
d λt
or {e P2 (t) = λ (λ t)
dt
λ ( λt ) t ( λt ) 2
or eλ t P2 (t) = + C = + C
2! 2!
where C is the constant of integration and its value is C = 0, for t = 0 and P2 (0) = 0. Thus,
( λt ) 2 ( λt ) 2 – λ t
eλ t P2 (t) = or P2 (t) = e
2! 2!
In general, we have,
( λt ) n − λ t
Pn (t) = e for n = 0, 1, 2, . . .
n!
This general solution for Pn (t) indicates the number of customers in the system at a time t before the start of the
service facility and it follows Poisson distribution with mean and variance equal to λt. The expected or mean number
of customers does not depend on the service time and therefore this solution holds good irrespective of the nature
of service that will be provided to the waiting customers in the system.

Remark The linear first order differential equation of the form

d
{ f (x)} + φ (x) ⋅ f (x) = P(x)
dx
has the solution

{ z }
f (x) = C exp − φ ( x ) dx + exp − φ ( x ) dx { z } LNM { z exp { z φ ( x ) dx } P ( x )} dx OQP
where C is a constant.
Alternative Method Equations (3) and (4) can also be solved by using the probability generating function approach.
Defining the probability generating function of Pn (t) as follows:
∞
G (z, t) = ∑ Pn ( t ) z n ; |z| ≤ 1
n=0

Also
d
{Gn (z, t)} =
d RS ∑ ∞
Pn ( t ) z n
UV = ∞
∑
d
P (t ) z n
dt dt T n=0 W n=0 dt n
∞
or G' (z, t) = ∑ Pn′ ( t ) z n
n=0
608 Operations Research: Theory and Applications

Multiplying both sides of Eq. (3) by zn and summing over the appropriate range of n, we get:
∞ ∞ ∞
∑ Pn′ ( t ) z n = − λ ∑ Pn ( t ) z n + λ ∑ Pn − 1 ( t ) z n (11)
n=0 n =1 n =1
Adding Eqs (4) and (11), we obtain:
∞ ∞ ∞
∑ Pn′ ( t ) z n = − λ ∑ Pn ( t ) z n + λ ∑ Pn − 1 ( t ) z n
n=0 n=0 n=0
or G' (z, t) = – λ G (z, t) + λ z G (z, t)
G ′ ( z, t )
or = λ (z – 1)
G ( z, t )
d
or { log G (z, t) } = λ (z – 1)
dt
Integrating both sides of this differential equation, we get:
log G (z, t ) = λ (z – 1) t + C (12)
where C is the constant of integration and its value can be obtained by using initial condition, i.e. set t = 0. By doing
so we get
log G (z, 0) = C for t = 0
∞ ∞
n
But G (z, 0) = ∑ Pn ( 0 ) z n = P0 (0) + ∑ z ⋅ Pn ( 0 ) = 1
n=0 n =1
since Pn (0) = 0, for n = 1. Thus C = log G (z, 0) = log 1 = 0. Hence, Eq. (12) reduces to the form
log G (z, t) = λ (z – 1) t or G (z, t) = eλ (z – 1) t
(13)
From generating function at z = 0, we obtain
dn
{G ( z , t )} = n! Pn (t)
dz n

or Pn (t) =
1 dn RS G ( z, t )
UV or P0 (t) = [G (z, t)]z = 0 = e– λ t
T
n ! dz n W
using the result Eq. (13) for n = 0 and z = 0. Similarly, at z = 0

RS d G ( z, t )UV = ne ( λ t )s =
λte
λ ( z − 1) t
−λ t
P1 (t) =
T dz W 1!
1 Rd
S G ( z, t )UVW = 2! e
2
2 ( λ t ) −λt
P (t) =
2
2! T dz 2
.
.
.
(λ t )n − λ t
Pn (t) = e , n = 0, 1, 2, . . .
n!
which is the same as derived earlier.
Theorem 16.2 (Distribution of Interarrival Times) If the arrival process follows the Poisson distribution,
n
(λ t )
Pn (t) = ⋅ e − λ t , n = 0, 1, 2, . . . (14)
n!
then an associated random variable defined as the interarrival time T follows the exponential distribution f (t) =
λ e– λ t and vice versa.
Proof Let T be the interarrival time, having a distribution function F(t). If there is no customer in the system at
time t = 0, then we have
F (t) = Prob (T ≤ t), probability that T takes on a value ≤ t.
= 1 – Prob (T > t) = 1 – Prob (no customers arrive during t)
= 1 – P0 (t) = 1 – e– λ t ; t ≥ 0
Differentiating both sides with respect to t, we get f (t) = F' (t) = λe– λ t, which is an exponential distribution. Here
f (t) is the probability density function for T.
 Queuing Theory 609

The expected (or mean) time of first arrival is given by

z z
∞ ∞
1
E (T ) = t f ( t ) dt = λ t e − λ t dt = (15)
λ
0 0
where λ is the mean arrival rate. Thus its variance would be 1/λ2.
Corollary The Markovian property of interarrival times states that the probability that a customer, currently in
service, is completed at sometime t is independent of how long he has already been in service. That is
Prob {T ≥ t1 such that T ≥ t0} = Prob {0 ≤ T ≤ t1 – t0}
where T is the time between successive arrivals.
Proof Since the left-hand side of Markovian property shows the conditional probability, it can be expressed as:
Prob(T ≥ t1 ) ∩ Prob (T ≥ t0)
Prob {T ≥ t1 | T ≥ t0 } =
Prob (T ≥ t0 )

z
t1
λ e − λ t dt
t0 − ( e − λ t1 − e − λ t 0 )
= ∞ = = 1 – e– λ (t1 – t0) (16)

z
t0
λe −λt
dt
e − λ t0

This is because of the reason that the interarrival times are exponentially distributed. Since

z
t1 − t 0
Prob (0 ≤ T ≤ t1 – t0) = λ e − λ t dt = 1 − e − λ ( t1 − t 0 ) (17)
0
therefore it is concluded that, Prob {T ≥ t1 | T ≥ t0} = Prob {0 ≤ T ≤ t1 – t0}

Distribution of Departures (Pure Death Process)

Sometimes a situation may arise when no additional customer joins the system while the service is continued for those
who are in queue. Let, at time t = 0 (e.g. closing hour) there be N ≥ 1 customers in the system. It is clear that the
service will be provided at the rate of µ. The number of customers in the system at time t ≥ 0 is equal to N minus
total departures up to time t. The distribution of departures can be obtained with the help of the following basic axioms.
(i) Probability of the departure during time ∆ t is µ ∆t.
(ii) Probability of more than one departure between time t and t + ∆t is negligible.
(iii) The number of departures in non-overlapping intervals are statistically independent, i.e. the process has
independent arrivals.
Let us define the terms that would be helpful in the development of various queuing models.
µ ∆t = probability that a customer in service at time t will complete service during time ∆t.
1 – µ ∆t = probability that the customer in service at time t will not complete service during
time ∆t.
For small time interval ∆t > 0, µ ∆t gives probability of one departure during time ∆t. Using the same arguments as
the one used in pure birth process case, the term 0 (∆t)2 → 0 as ∆t → 0 and hence it can be neglected. Hence the
differential-difference equations for this can be obtained as given below:
PN (t + ∆t) = PN (t) {1 – µ ∆t} ;n = N , t ≥ 0
Pn (t + ∆t) = Pn (t) {1 – µ ∆t} + Pn + 1 (t) µ ∆t ;1 ≤ n ≤ N – 1 , t ≥ 0
P0 (t + ∆t) = P0 (t) + P1 (t) {µ ∆t} ;n = 0 , t = 0
Rearranging the terms and taking the limits as ∆ t → 0, we get:
PN′ ( t ) = – µ PN (t) ; n = N,
Pn′ ( t ) = – µ Pn (t) + µ Pn + 1 (t) ; 0 ≤ n ≤ N – 1,
P0′ ( t ) = µ P1 (t) ; n = 0,
The solution of these equations with the initial conditions:

Pn (0) =
RS1 ; n= N ≠0
T0 ; n≠ N
610 Operations Research: Theory and Applications

can be obtained in the same manner as in the case of pure birth process. The solution to these equations, so obtained,
is:
R| (µ t ) e N−n −µ t
; 1 ≤ n ≤ N, t ≥ 0
|| ( N − n)!N
P (t) = S1 − ∑ P ( t ) ; n = 0, t≥0
n
||0 n =1
n

; n ≥ N + 1, t ≥ 0
|T
Distribution of Service Times
The probability density function s(t) of service time T = t is given by
d µ e −µ t 0; ≤ t ≤ ∞
s (t) = {S (t )} = 
dt 0 o;therwise
The proof for this is exactly on the same as discussed earlier. This shows that service times follows exponential
distribution with mean 1/µ and variance 1/µ2.

APPENDIX 16.B: ERLANGIAN SERVICE TIME DISTRIBUTION

WITH K-PHASES
Step 1: Obtain system of differential difference equations
Let Pn (t) be the probability that there are n phases in the system at time t. Now the difference equations for this
model are:
Pn (t + ∆t) = – (λ + k µ) Pn (t) ∆t + k µ Pn + 1 (t) ∆t + Pn – k (t) ∆t + Pn (t)
+ terms containing (∆t)2 ; n ≥ 1.
and P0 (t + ∆t) = P0 (t) (1 – λ ∆t) + P1 (t) k µ ∆t ; n = 0

Step 2: Obtain the system of steady-state equations

The system of steady-state difference equations can be obtained by using the usual procedure.
– ( λ + k µ) Pn + k µ Pn + 1 + λ Pn – k = 0 ; n ≥ 0
and – λ P0 + k µ P 1 = 0 ; n = 0
Let ρ = λ/k µ, and dividing each equation by kµ, the system of equations reduces to the form:
(1 + ρ) Pn = ρ Pn – k + Pn + 1 ; n ≥ 1 (18)
and P 1 = ρ P0 ; n= 0 (19)

Step 3: Solve the system of equations

For solving the given system of difference equations we make use of the method of generating function (GF) defined
as
∞
G (x) = ∑ Pn x n ; |x| ≤ 1 (20)
n=0
Multiplying Eq. (18) by xn and summing over the range from n = 1 to ∞, we get:
∞ ∞ ∞
(1 + ρ) ∑ Pn x n = ρ ∑ Pn − k x n + ∑ Pn + 1 x n (21)
n =1 n=1 n =1
Adding ρ P0 on the left side of Eq. (21) and P1 (= ρ P0 ) on the right of Eq. (21),we get:
∞ ∞ ∞
(1 + ρ) ∑ Pn x n + ρ P0 = P1 + ρ ∑ Pn − k x n + ∑ Pn + 1 x n
n =1 n =1 n =1

LM ∞ OP – P ∞ LM
Pn − k x n + P1 + ∑
∞
Pn + 1 x n
OP
(1 + ρ) P0 + ∑ Pn x n = ρ ∑
MN n =1 PQ 0
n=1 MN n =1 PQ
∞ ∞ 1 ∞
(1 + ρ) ∑ Pn x n – P0 = ρ ∑ Pn − k x n + ∑ P xn +1
n=0 n=k x n = 0 n +1
 Queuing Theory 611

Since Pn –k = 0 for n – k = 0, therefore we have:

∞ ∞ 1 ∞
(1 + ρ) ∑ Pn x n – P0 = ρ ∑ Pj x j + k + ∑ P x i; for n – k = j; n + 1 = i
n=0 j=0 x i =1 i

= ρx
k
∑ pjx j +
∞ 1 LM ∑ ∞
Pi x i − P0
OP
j=0 x MN i=0 PQ
1
or (1 + ρ) G (x) – P0 = ρ x k G (x) + [G (x) – P0]
x
P0 (1 − x ) LM F 1 − x I OP k
−1

or G (x) =
(1 − x ) − ρ x (1 − x k )
= P0 1 − ρx
MN GH 1 − x JK PQ ; |x| ≤ 1

= P0 ∑
∞
( x ρ) n FG 1 − x IJ k n

n=0 H 1− x K (Expansion by binomial theorem) (22)

FG 1 − x IJ
k
= 1 + x + x2 + . . . + xk – 1
Since
H 1− x K ,

∞
therefore G (x) = P0 ∑ ( x ρ ) n (1 + x + x2 + . . . + xk – 1 n
)
n=0
∞
= P0 ∑ ρ n ( x + x 2 + x 3 + . . . + x k ) n
n=0
To find value of P0 and Pn, let us first substitute x = 1, we obtain

n n
∞
G (1) = P0 ∑ ρ k = P0
FG 1 IJ (Sum of infinite GP)
n=0 H 1 − kρ K
for x = 1, Eq. (20) gives
∞
G (1) = ∑ Pn = 1
n=0

1 = P0
FG 1 IJ
Thus
H 1 − kρ K or P0 = 1 – k ρ

Substituting the value of P0 in Eq. (22), we get:

∞
G (x) = (1 – k ρ) ∑ ( x ρ) n (1 − x k ) n (1 − x ) − n (23)
n=0

But d1 − x i
k n = 1 – n C1 x k + n C2 x k
2
d i
+ . . . + ( −1) n n C n x k
n
d i
n
t n tk
= ∑ ( −1) Ct x
t=0
∞
and (1 – x)– n = ∑ ( −1) i − n Ci x i
i=0
∞
= ∑ (−1) 2i ( n + i − 1) Ci x i ; −n
Ci = (– 1)i (n + i – 1) Ci
i=0
Therefore, Eq. (23) reduces to
LMR ∑ n UV RS ∑
∞ UVOP ; (– 1)
MNST
∞ n + i − 1C
G (x) = (1 – k ρ) ∑ ( xρ) n ( −1) t nC x tk xi 2i
n=0 t=0
t
WTi=0
i
WPQ =1

or
∞
∑ Pn x n = (1 – k ρ) ∑ ρ n
∞ LM ∑ n ∞
∑ ( −1) t n
Ct n + i −1
Ci x i + tk + n (24)
n=0 n=0 N t=0 i=0
Comparing the coefficient of xn from both sides of Eq. (24), we have
Pn = (1 – k ρ) ∑ ρ n ( −1) t n Ct n + i − 1Ci
n, t , i
612 Operations Research: Theory and Applications

∞
Important result To determine the expected number of phases in the system: Ls (k) = ∑ nPn , consider one
n=0
of the difference equations as
(1 + ρ) Pn = ρPn – k + Pn + 1; n ≥ 1
Multiplying both sides by n2 and taking summation from n = 1 to ∞, we get
∞ ∞ ∞
2 2
(1 + ρ) ∑ n 2 Pn = ρ ∑ n Pn − k + ∑ n Pn + 1
n =1 n =1 n =1

∞ ∞ ∞
2 2
or (1 + ρ) ∑ n 2 Pn = ρ ∑ n Pn − k + ∑ n Pn + 1
n =1 n=k n =1
Suppose n + k ≡ n and n – 1 ≡ n, then RHS becomes
∞ ∞
2 2
RHS = ρ ∑ (n + k ) Pn + ∑ (n − 1) Pn
n=0 n =1
∞
2  ∞ 2 
= ρ ∑ (n + k ) Pn +  ∑ (n − 1) Pn − P0 
n=0  n = 0 
∞
2 2
= ∑ [ρ (n + k ) + (n − 1) ] Pn − P0
n=0
∞
2 2 2
= ∑ [ρ ( n + k + 2 nk ) + (n + 1 − 2 n)] Pn − P0
n=0
∞
2 2
= ∑ [(1 + ρ)n − 2n (1 − k ρ) + (ρ k + 1)] Pn − P0
n=0
∞ ∞ ∞
2 2
= (1 + ρ) ∑ n Pn − 2(1 − kρ) ∑ nPn + 2(1 + ρ k ) ∑ Pn − P0
n=0 n=0 n=0
Equating RHS and LHS, we get
∞
2 (1 – ρ k) ∑ nPn = ρk2 + 1 – P0
n=0
∞ ρ k 2 + 1 − P0 ρ k 2 + 1 − (1 − ρ k )
or L s (k) = ∑ nPn = =
n=0 2(1 − ρ k ) 2(1 − ρ k )
k (1 + k ) ρ (1 + k ) λ
or L s (k) = ⋅ = ⋅ ; ρ = λ /µ k
2 1 − ρk 2 µ−λ

Remark Other results can be derived with the help of Ls (k) as shown in the main text.