CLUSTER UNIVERSITY OF JAMMU

Jammu,J&K-180001

A Project Report
On
”QUEUEING THEORY”
Submitted in partial fulfilment of
the requirements for the award of the degree of
INTEGRATED M.Sc.
In
MATHEMATICS
Submitted by

ETISHA RAJPUT USN1

Under the Guidance of

DR. NITIN KUMAR PhD, M.Sc.

Assistant Proffessor

DEPARTMENT OF MATHEMATICS
INDIAN INSTITUTE OF TECHNOLOGY
JAMMU-181221
2023-24
Contents
1 INTRODUCTION 3
1.1 Origin of Queueing theory . . . . . . . . . . . . . . . . . . 3
1.2 Characteristics of Queueing system . . . . . . . . . . . . 5
1.3 Kendall Notation . . . . . . . . . . . . . . . . . . . . . . . . 8

2 MODELS 10
2.1 M/M/1 Model . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2 GI/M/1 MODEL . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 GI X /M/1 MODEL . . . . . . . . . . . . . . . . . . . . . . . 20

2
1 INTRODUCTION
Waiting in a line/queue is a common problem in everyday life, whether we’re
stuck in traffic, at a bus stop, we wait on hold for an operator to pick up our
telephone calls, we wait in line at supermarket, banks, post office, airport
etc. The discomfort of waiting affects both service seekers and providers. But
why do we have to wait? It happens when our demands does not match by
the available facilities provided by the server. This is because of the factors
like limited resources, high demand, unpredictable delays and many more.
Individuals or entities seeking service are often referred to as ‘customers’ and
the place or entity delivering the service is called the ‘server’. In general, a
queue is formed when there are customers in need of some sort of services
and the queueing problem arises when a group of customers arrives randomly
to receive some service. The term ‘customer’ is used in a general sense. It
may be a person, machine, vehicle or anything else which requires service.
Queueing theory is a branch of mathematics dedicated to studying queues.
It provides a framework for understanding and optimizing systems involving
waiting lines. Many real-world systems can be modeled using the concept
of a ‘queue’ or ‘waiting line.’ A queue is essentially a line of customers or
objects waiting for service.It can also be classified as a part of stochastic
processes. The theory attempts to minimizie the cost of waiting without
excessively increasing the cost of servicing. Sometimes Queueing theory, also
referred as the ‘waiting line theory.’ It is significant because it helps to de-
scribe queue charateristics such as average wait time and provide tool for
queue optimization. Queueing theory finds applications in various fields, in-
cluding telecommunications, transportation, healthcare, computer science,
retail, and the service industry. It was developed to create models that pre-
dict the behavior of systems handling randomly occurring demands.

1.1 Origin of Queueing theory

The history of Queueing theory started almost 100 years ago. It was de-
veloped by Anger Krarup Erlang (1878-1929), a Danish mathematician,
1.1 Origin of Queueing theory 4

statistician and engineer, in the early 20th century. Erlang worked for the
Copenhagen Telephone Exchange, where he aimed to improve the company
processes. He wanted to know how many telephone lines were needed so that
callers didn’t have to wait too long to get connected and how many telephone
operators were required to handle a specific number of calls effectively. In
1909 Erlang published his first work, “The Theory of Probabilities and Tele-
phone Conversation”. In this work he showed that when telephone calls are
made randomly follow Poisson’s law of distribution.
In 1917, Erlang published his most crucial work “Solution of some Problems
in the Theory of Probabilities of Significance in Automatic Telephone Ex-
change”. In this paper, he introduced important formulas that have become
fundamental in the field of telephone traffic theory. His pioneering work in-
troduced fundamental concepts like arrival rates, service times, and waiting
times, which became the building blocks of queueing models.
In the 1950s, where D.G. Kendall played a significant role by standardizing
the notation system, giving birth to Kendall’s notation. This notation sys-
tem, including designations like M/M/1 and M/G/1, brought a revolutionary
approach to describing queueing systems, simplifying their classification and
analysis based on the distinct characteristics of arrival and service times.
In the 1960s, John Jackson’s expansion of queueing theory into networked
systems was a crucial development. This extension enabled the analysis of
complex situations where customers move between queues, which was essen-
tial for understanding diverse systems like computer networks and manufac-
turing processes.
The 1970s marked a turning point with the arrival of computers, as numerical
methods and simulation techniques were introduced, enabling the analysis of
increasingly complex queueing models and the exploration of various param-
eters. This computational progress broadened the practical applications of
queueing theory.
During the 1980s and 1990s, researchers started using more realistic models
in queueing theory. They began including service times that didn’t follow
the typical exponential pattern, which is more common in real-world systems.
They introduced models that considered different service time distributions
1.2 Characteristics of Queueing system 5

like Erlang, hyperexponential, and general distributions. This made the mod-
els better at representing actual systems.
In the 1990s, people realized the importance of priority queuing and multi-
server models in queueing theory. These models allowed us to handle sit-
uations where there are different types of customers, multiple servers, and
varying priorities. This was necessary because systems were becoming more
complex, and these models helped us manage that complexity better. In the
21st century, queueing models became very important in real-time systems.
They helped in making quick decisions and managing resources in areas like
finance, healthcare, and online shopping. Moreover, when combined with
data science and machine learning, they opened up new possibilities for pre-
dicting outcomes, optimizing processes, and automating tasks in various in-
dustries.

1.2 Characteristics of Queueing system

Generally, queueing models are used when we can’t predict when customer
will arrive or how long they will take to be served because arrival-departure
processes are random. So, these models help to reduce waiting time and costs
by finding the right balance between service capacity and waiting time.

Mostly six basic characteristics of a queueing system are: (1) arrival pro-
cess of customers, (2) service process of servers, (3) queue discipline, (4)
system capacity, (5) number of servers (channel) and (6) number of service
stages. Certainly, the first five characteristics are crucial for understanding
and managing the fundamentals of most queueing systems. We will discuss
these characteristics in details.

ˆ Arrival Process

In usual queueing system, the process of arrivals is stochastic. The arrival

process describes the manner in which the customers arrive and join the
queueing system. It is not possible to observe and control the actual moment
1.2 Characteristics of Queueing system 6

of arrival of a customer for service. Thus the intensity of arrivals in a system

is measured by the average number of arrivals per unit of time, known as the
arrival rate or the average time between the successive arrivals, known as the
interarrival time. These two measures help us to understand and analyze the
flow of arrivals in various processes or systems. The way arrivals happen is
commonly categorized by the following patterns:

ˆ Unit or batch arrivals (customers arrive one at at time or in groups)

ˆ Renewal arrivals (times between arrivals are independent random vari-

ables)

ˆ Stationary arrivals (parameters of arrivals process do not change over

time )

ˆ Non-stationary or Independent increments (arrival occurs indepen-

dently without being influenced by previous arrivals)

It is also very important to know the behaviour of a arrival/customer upon

entering the system. A customer may decide to wait until served no matter
how long the queue becomes. Such a customer is called patient customer. On
the other hand, if the queue is deemed long, the customer may decide not to
enter the system. Such type of behaviour is called balking. If a customer may
enter the queue, and after waiting for sometime decides to leave the system.
Such type of behaviour is called reneging. Further, a customer who enters
a queue and there are two or more parallel waiting lines, so customers may
switch from one line to another. Such type of behaviour is called jockeying.
These three situations are all examples of queues with impatient customers.

ˆ Service Process

The mode of service is described by looking at how many customers are served
in a given amount of time or the time gap between serving each customer.
This information is typically represented using a probability distribution.
Service can be given either to a single customer such as a patient in a hospi-
tal, or to a batch of customers such as a busload of tourists on a tour. In some
1.2 Characteristics of Queueing system 7

cases, multiple customers can be served at the same time by a single server.
This happens in situations like when a computer uses parallel processing,
when a group of sightseers are guided together, or when people board a train
simultaneously. The service process may depend on the number of customers
waiting for service. Such situtation is referred to as state-dependent service.
Service rate may or may not depend on the congestion in the system.
Service, similar to customer arrivals, can be either stationary or non-
stationary. For example, as people gain experience, the service can become
more efficient. It’s important to note that this time-related change is not
connected to how busy the system is but rather how long it’s been running.
On the other hand, the state-dependent aspect of service is not related to
how long the system has been running but rather to how many customers
are currently in the system. Of course, a queueing system can exhibit both
non-stationary and state-dependent behavior.
Even when the service rate is high, there is still a good chance that some cus-
tomers will experience delays while waiting in line. Usually, customers arrive
and are served at unpredictable times, which means the queue length doesn’t
follow a specific pattern unless both arrivals and service are deterministic.
Therefore, the probability distribution for queue lengths is influenced by two
separate processes: customer arrivals and service. These processes are often
assumed to be independent of each other, although not always.

ˆ Queue Discipline

Queue discipline refers to the rule by which customers are selected for service
when a queue has formed. In our day to day life the most common discipline
that can be observed is first come, first served (FCFS) or first in, first out
(FIFO). Another discipline is last come, first served (LCFS), which is often
used in storage systems where the last arrived item may be served first as
it is easier to reach the nearest items than the item which is stored in the
beginning. Service-in-random-order (SIRO), which means server chooses one
of the customer at random order for service.
Another way to handle queues is by considering the priority of customers.
Customers with higher priorities are selected first for service than with lower
1.3 Kendall Notation 8

priorities, irrespective of their time of arrival in the system. There are two
general situations in priority disciplines. The first one is called preemptive
queues, the customer with the higher priority arrives, is allowed to enter ser-
vice immediately even if a customer with lower priority is already in service
(such as in hospital emergency room). The lower priority customer service is
temporarily paused and will resume after the higher priority one is taken care
of. This is called preemption. The second one is called the non-preemptive
queues, the higher priority customer waits until the service of the presently
customer is done, even though the present customer has a lower priority.

ˆ System Capacity

System capacity refers to the maximum number of customers that a system

can accommodate at one time, due to physical limitations. When the queue
reaches a certain length and there is no more space, new customers have to
wait until someone finishes their service before they can join. These situa-
tions are termed as “finite queueing situations” because there is a definite
limit to the maximum system size. Such system can be considered as those
where arrival customer is compelled to balk (forced balking) if the queue has
reached its maximum capacity.

ˆ Number of Servers

1.3 Kendall Notation

In Queueing theory Kendall Notation is the standard system used to clas-
sify different types of queueing models. In other words we can say that
Kendall Notation is a system of notation used to identify and describe
the key characteristics of a queuing model. It’s a way to represent impor-
tant details about how queues work, such as arrival rates, service rates, and
the number of service channels or servers. The notation introduced by D.G
Kendall in 1953. A queueing model is described by a series of symbols and
slashes such as A/B/C/D/E ,where A indicates the interarrival-time distri-
bution, B the service pattern as described by the probability distribution
1.3 Kendall Notation 9

for service time, C the number of parallel service channels, D the restriction
on system capacity or maximum number of customers allowed in the system
and E the queue discipline. Some standard symbols for these characteristics
are presented in Table 1.1. For example the notation M/M/1 means that
the time between arrivals is Markovian (M) i.e the inter-arrival time follows
an exponential distribution of parameter λ where as second M means that
the service time is Markovian that follows an exponential distribution of pa-
rameter µ and 1 denotes the single server. This notation system helps in
analyzing and comparing different queuing systems and understanding how
efficiently they operate. It provides a standardized way to communicate and
study various aspects of queues and waiting lines. It is widely used in re-
search and practice to study the behaviour of queueing systems and to make
informed decisions about their design and operation.
2 MODELS

2.1 M/M/1 Model

The M/M/1 model is a simple queueing theory model in which M stands
for Poisson arrivals, M for exponential service times, and 1 for one server.
1 1
In this model, λ is the arrival rate and µ is the service rate then λ
and µ
show mean arrival time and mean service time correspondingly.For the given
model,arrivals follow Poisson distribution and service time follows exponen-
tial distrubtion.Here arrival times and service times are generally indepen-
dent of one another.
Let N(t) be the number of customer in the system and pn (t) is the probability
that there are n customer in the system at time t.
We define
pn (t) = P [N (t) = n], n≥0

and in steady state

pn = lim pn (t), n≥0
t→∞

Now we have,

p0 (t + ∆t) = p1 (t){(µ∆t + o(∆t))(1 − λ∆t + o(∆t))}

+ p0 (t){(1 − λ∆t + o(∆t))(1 − µ∆t + o(∆t))}
= p1 (t)(µ∆t + o(∆t)) + p0 (t)(−λ∆t + o(∆t))

pn (t + ∆t) = pn−1 (t){(λ∆t + o(∆t))(1 − µ∆t + o(∆t))}

+ pn+1 (t){µ∆t + o(∆t)(1 − λ∆t + o(∆t))}
+ pn (t){(1 − λ∆t + o(∆t))(1 − µ∆t + o(∆t))}
= pn−1 (t)(λ∆t + o(∆t)) + pn+1 (t)(µ∆t + o(∆t))
+ pn (t)(1 − µ∆t + o(∆t) − λ∆t)

Using Taylor series expansion for 2 variables,

d
pn (t + ∆t) = pn (t) + ∆t pn (t) + o(∆t), n≥0
dt

10
2.1 M/M/1 Model 11

Next, we have
d
p0 (t) + ∆t p0 (t) + o(∆t) = p1 (t)(µ∆t + o(∆t)) + p0 (t)(−λ∆t + o(∆t))
dt
(1)

d
pn (t) + ∆t pn (t) + o(∆t) = pn−1 (t)(λ∆t + o(∆t)) + pn+1 (t)(µ∆t + o(∆t))
dt
+ pn (t)(1 − λ∆t − µ∆t + o(∆t)) (2)

Dividing both side of (1) and (2) by ∆t and taking lim we get,
∆t→0

d
p0 (t) = µp1 (t) + (−λ)p0 (t)
dt
d
pn (t) = λpn−1 (t) + µpn+1 (t) + (−λ − µ)pn (t); n≥1
dt
We define pn = lim pn (t)
t→∞
Taking t → ∞, we get

0 = µp1 − λpo (3)

0 = λpn−1 + µpn+1 + (−λ − µ)pn ; n≥1 (4)
From (3) we get,
λpo
p1 =
µ

Put n=1 in (4) we get

λpo + µp2 + (−λ − µ)p1 = 0

λpo
µp2 = (λ + µ) − λpo
µ
λ2 po + λµpo − λµp0
µp2 =
µ
2
λ po
p2 =
µ2

Next put n=2 in (4) we get

λp1 + µp3 + (−λ − µ)p2 = 0

2.1 M/M/1 Model 12

λ2 po  λp 
o
µp3 = (λ + µ) 2 − λ
µ µ
3 2 2
λ po + λ µpo λ po
µp3 = −
µ2 µ
λ po + λ µpo − λ µpo
3 2 2
µp3 =
µ2
λ3 po
p3 = 3
µ
.
.
.
λn po
pn = ; n≥1
µn

We know that,

∞
X
pn = 1
n=0
X∞
p0 + pn = 1
n=1
∞  
X λ n
p0 + p0 = 1
n=1
µ
∞ 
 X λ n 
p0 1 + = 1
n=1
µ
 λ 
µ
p0 1 + λ
= 1
1− µ
 λ 
p0 1 + = 1
µ−λ
 µ 
p0 = 1
µ−λ
µ−λ
p0 =
µ
λ
p0 = 1 −
µ
2.1 M/M/1 Model 13

Now,

∞
X
Mean(L) = npn
n=1
X∞  λ n
L = n p0
n=1
µ
λ
µ
= p0
(1 − µλ )2
λ
µ
= p0
( µ−λ
µ
)2
2
λ (µ) λ
= 2
(1 − )
µ (µ − λ) µ
λ
=
µ−λ
ρ λ
L = , where ρ=
1−ρ µ

and

∞
X
Lq = (n − 1)pn
n=1
X∞  λ n
= (n − 1) p0
n=1
µ
∞  λ n ∞  
X X λ n
= n p0 − p0
n=1
µ n=1
µ
λ
ρ  
µ
= − p 0
1−ρ 1 − µλ
ρ λ  λ
= − 1−
1−ρ µ−λ µ
ρ λ
= −
1−ρ µ
ρ
= −ρ
1−ρ
ρ2
Lq =
1−ρ
2.1 M/M/1 Model 14

Also,
∞
X ∞
hX i2
2
Variance(V) = n pn − npn
n=1 n=1
∞
X  λ n  ρ 2
2
= n p0 −
n=1
µ 1−ρ
X∞  ρ 2
= n2 (ρ)2 p0 −
n=1
1−ρ
ρ(1 + ρ)  ρ 2
= p 0 −
(1 − ρ)3 1−ρ
2
ρ+ρ  ρ 2
= (1 − ρ) −
(1 − ρ)3 1−ρ
2 2
ρ+ρ ρ
= 2
−
(1 − ρ) (1 − ρ)2
ρ
=
(1 − ρ)2
2.2 GI/M/1 MODEL 15

2.2 GI/M/1 MODEL

In this model, both the inter-arrival times and service times can follow ar-
bitrary probability distributions. Here GI denotes the general distribution,
M denotes the exponential distribution for service times and 1 for single
server.The arrival of a GI/M/1 queue are given by a renewal process, where
this renewal process must be specifically by a poisson process.
We specify the states of the system at time t as (N(t),U(t)) where
• N(t)=Number of customers in the system
• U(t)=Remaining inter-arrival time of the next batch. We define

pn (t) = P [N (t) = n], n≥0

pn (t, u)∂u = P [N (t) = n, u < U (t) ≤ u + ∂u], u ≥ 0, n ≥ 0

and in steady-state

pn = lim pn (t) and pn (u) = lim pn (t, u), n≥0

t→∞ t→∞

p0 (t + ∆t, u − ∆u) = p1 (t, u){µ∆t + o(∆t)} + p0 (t, u){1 − µ∆t + o(∆t)}

= p1 (t, u){µ∆t + o(∆t)} + p0 (t, u){1 + o(∆t)}

pn (t + ∆t, u − ∆u) = pn−1 (t, u){(a(u)∆u + o(∆u))(1 − µ∆t + o(∆t))}

+ pn+1 (t, u){µ∆t + o∆t} + pn {1 − µ∆t + o(∆t)}
= pn−1 (t, u){(a(u)∆u + o(∆u))} + pn+1 (t, u){µ∆t + o∆t}
+ pn {1 − µ∆t + o(∆t)}

Using Taylor series expansion for two variable,

∂ ∂
pn (t + ∆t, u − ∆u) = pn (t, u) + ∆t pn (t, u) − ∆u pn (t, u) + o(∆t), n≥0
∂t ∂u
Next we have,
∂ ∂
p0 (t, u) + ∆t p0 (t, u) − ∆u p0 (t, u) + o(∆t) = p1 (t, u){µ∆t + o(∆t)}
∂t ∂u
+p0 (t, u){1 + o(∆t)} (5)
∂ ∂
pn (t, u) + ∆t pn (t, u) − ∆u pn (t, u) + o(∆t) = pn−1 (t, u){(a(u)∆u + o(∆u))
∂t ∂u
+pn+1 (t, u){µ∆t + o∆t}
+pn {1 − µ∆t + o(∆t)} (6)
2.2 GI/M/1 MODEL 16

Dividing both side of (5) and (6) by ∆t and taking lim we get,
∆t→0

∂ ∂ 
− p0 (t, u) = µp1 (t, u)
∂t ∂u
∂ ∂ 
− pn (t, u) = −µpn (t, u) + a(u)pn−1 (t, 0) + µpn+1 (t, u)
∂t ∂u
Taking t → ∞
d
− p0 (u) = µp1 (u) (7)
du
d
− pn (u) = −µpn (u) + a(u)pn−1 (0) + µpn+1 (u) ;n ≥ 1 (8)
du
R∞
Multiply (7) and (8) by e−su and u=0 , we obtained the tranformed equation as,

−sp∗0 (s) + p0 (0) = µp∗1 (s) (9)

−sp∗n (s) + pn (0) = −µp∗n (s) + A∗ (s)pn−1 (0) + µp∗n+1 (s), n ≥ 1 (10)

We define difference operator as,

Dp∗n (s) = p∗n+1 (s)

Dpn (0) = pn+1 (0)

Now applying difference operator in (10),we have

(−s + µ − µD)p∗n (s) = (A∗ (s) − D)pn−1 (0), n≥1

Now put s = µ − µD we get,

(A∗ (µ − µD) − D)pn−1 (0) = 0, n≥1

(A∗ (µ − µD) − D)pn (0) = 0, n≥0 (11)

The auxiliary equation corresponding to (11) is given by,

A∗ (µ − µz) − z = 0 (12)

Equation (12) has one root inside the unit circle,say α.

The solution of (12) is given by

pn (0) = cαn , n≥0

2.2 GI/M/1 MODEL 17

So we have,

(A∗ (s) − α)cαn−1

p∗n (s) = , n≥1
−s + µ − µα
Setting s → 0,we get

(1 − α)cαn−1
pn = p∗n (0) = , n≥1
µ − µα
and
pn (0)
p−
n =
λ
cαn
=
λ
Now adding (9) and (10) we get,
∞
X ∞
X ∞
X ∞
X ∞
X
−s p∗n (s) + pn (0) = −µ p∗n (s) +µ p∗n+1 (s) ∗
+ A (s) pn−1 (0)
n=0 n=0 n=1 n=0 n=1
X∞ X∞ X∞ X∞ ∞
X
−s p∗n (s) + pn (0) = −µ p∗n (s) + µ p∗n (s) + A∗ (s) pn (0)
n=0 n=0 n=1 n=1 n=0
X∞ ∞
 A∗ (s) − 1  X
− p∗n (s) = pn (0)
n=0
s n=0

Let s → 0 and using A∗ (0) = 1

Also define
pn = p∗n (0)
∞
X X
− pn = A∗(1) (s) s = 0 n = 0∞ pn (0)
n=0
∞
1X
−1 = − pn (0)
λ n=0
∞
X
pn (0) = λ
n=0
∞
X
cαn = λ
n=0
2.2 GI/M/1 MODEL 18

1
c =λ
1−α
c = λ(1 − α)

Using the value of c in pn and p−

n we get,

λ(1 − α)αn−1
pn = , n≥1
µ
p− n
n = (1 − α)α , n≥0

We know that
∞
X
pn = 1
n=0
∞
X
p0 + pn = 1
n=1
∞
X λ(1 − α)αn−1
p0 + =1
n=1
µ
∞
λ(1 − α) X n
p0 = 1 − α
µα n=1

λ(1 − α)  α 
p0 = 1 −
µα 1−α
λ
p0 = 1 −
µ
Now,
∞
X
Mean(L) = npn
n=1
X∞  λ(1 − α)αn−1 
L= n
n=1
µ
∞
 λ(1 − α)  X
= nαn
µα n=1
 λ(1 − α)  α 
=
µα (1 − α)2
ρ
L=
1−α
2.2 GI/M/1 MODEL 19

Also
∞
X ∞
hX i2
2
Variance(V ) = n pn − npn
n=1 n=1
 λ(1 − α) ∞
X  ρ 2
= n2 αn −
µα n=1
1−α
 λ(1 − α)  α(1 − α)  α2
= −
µα (1 − α)3 (1 − α)2
ρ(1 + α) ρ2
= −
(1 − α)2 (1 − α)2
ρ(1 + α) − ρ2
V =
(1 − α)2
2.3 GI X /M/1 MODEL 20

2.3 GI X /M/1 MODEL

In this model GI X represents General inter-arrival time with batch arrival,
M stands for exponential service time and 1 for single server.The customer
arrive in group or batches of random size X with probability mass function
P (X = i) = gi ,i=1,2,...b,probability generating function G(z) = bi=1 gi z i
P
Pb
and the mean batch size ḡ = i=1 igi .Service time have an exponential
distribution with rate µ.
We specify the states of the system at time t as (N(t),U(t)) where
• N(t)=Number of customers in the system
• U(t)=Remaining inter-arrival time of the next batch.
We define
pn (t) = P [N (t) = n], n≥0

pn (t, u)∂u = P [N (t) = n, u < U (t) ≤ u + ∂u], u ≥ 0, n ≥ 0

and in steady-state

pn = lim pn (t) and pn (u) = lim pn (t, u), n≥0

t→∞ t→∞

p0 (t + ∆t, u − ∆u) = p1 (t, u){µ∆t + o(∆t)} + p0 (t, u){1 − µ∆t + o(∆t)}

= p1 (t, u){µ∆t + o(∆t)} + p0 (t, u){1 + o(∆t)}

pn (t + ∆t, u − ∆u) = pn−1 (t, u){λg1 ∆t + o(∆t))(1 − µ∆t + o(∆t))}

+ pn−2 (t, u){λg2 ∆t + o(∆t))(1 − µ∆t + O(∆t))} + ......
+ pn−b (t, u){λgb ∆t + o(∆t))(1 − µ∆t + o(∆t))}
+ pn+1 (t, u){µ∆t + o∆t} + pn {1 − µ∆t + o(∆t)}
b
X
= pn−i (t, u){λgi ∆t + o(∆t)} + pn+1 (t, u){µ∆t + o∆t}
i=1

+ pn {1 − µ∆t + o(∆t)}

Using Taylor series expansion for two variable,

∂ ∂
pn (t + ∆t, u − ∆u) = pn (t, u) + ∆t pn (t, u) − ∆u pn (t, u) + O(∆t), n≥0
∂t ∂u
2.3 GI X /M/1 MODEL 21

Next we have,
∂ ∂
p0 (t, u) + ∆t p0 (t, u) − ∆u p0 (t, u) + o(∆t) = p1 (t, u){µ∆t + o(∆t)}
∂t ∂u
+p0 (t, u){1 + o(∆t)} (13)

∂ ∂
pn (t, u) − ∆u pn (t, u) + o(∆t) = bi=1 pn−i (t, u){λgi ∆t + o(∆t)}
P
pn (t, u) + ∆t
∂t ∂u
+pn+1 (t, u){µ∆t + o(∆t)}
+pn {1 − µ∆t + o(∆t)} (14)

Dividing both side of (13) and (14 )by ∆t and taking lim we get,
∆t→0
∂ ∂ 
− p0 (t, u) = µp1 (t, u)
∂t ∂u
∂ b
∂  X
− pn (t, u) = −µpn (t, u) + a(u) pn−i (t, 0)gi + µpn+1 (t, u)
∂t ∂u i=1

Taking t → ∞
d
− p0 (u) = µp1 (u) (15)
du
n
d X
− pn (u) = −µpn (u) + a(u) pn−i (0)gi + µpn+1 (u), 1 ≤ n ≤ b − 1 (16)
du i=1
b
d X
− pn (u) = −µpn (u) + a(u) pn−i (0)gi + µpn+1 (u), n≥b (17)
du i=1
R∞
Multiplying equation (15)-(17) by e−su and u=0 , we get

sp∗0 (s) = p0 (0) − µp∗1 (s) (18)

n
X
sp∗n (s) = pn (0) + µp∗n (s) − µp∗n+1 (s) − A∗ (s) pn−i (0)gi , 1 ≤ n ≤ b − 1
i=1
(19)
b
X
sp∗n (s) = pn (0) + µp∗n (s) − µp∗n+1 (s) − A∗ (s) pn−i (0)gi , n≥b (20)
i=1

We define difference operator as,

Dp∗n (s) = p∗n+1 (s)

Dpn (0) = pn+1 (0)
2.3 GI X /M/1 MODEL 22

Now applying difference operator in (17),we have

 b
X 
(s − µ + µD)p∗n (s) b
= D − A (s) ∗
gi Db−1 pn−b (0), n≥b
i=1

Now put s = µ − µD we get,

 b
X 
b ∗
D − A (µ − µD) gi Db−1 pn−b (0) = 0, n≥b
i=1
 b
X 
b ∗
D − A (µ − µD) gi Db−1 pn (0) = 0, n≥0 (21)
i=1

The auxiliary equation corresponding to (21) is given by,

b
X
b ∗
z − A (µ − µz) gi z b−i = 0 (22)
i=1

Equation (22) has exactly b root inside the unit circle,say r1 , r2 .......rb and

b
X
pn (0) = ci rin ; n ≥ 0
i=1

Next, we have

b
X b
X b
X
(s − µ + µD)p∗n (s) = cj rjn − A (s) ∗
gi cj rjn−i
j=1 i=1 j=1

b n rb − A∗ (s) Pb gi rb−i o
j
X i=1 j
p∗n (s) = cj rjn−b , n≥b (23)
j=1
s − µ + µr j

For p∗n (s), 1 ≤ n ≤ b − 1, we seek a similar expression as given in (23)

Assume that
b n rb − A∗ (s) Pb gi rb−i o
j
X i=1 j
p∗n (s) = cj rjn−b is true for n = b − 1
j=1
s − µ + µr j
2.3 GI X /M/1 MODEL 23

From (16), we have

b n rb − A∗ (s) Pb gi rb−i o b n b
j
X i=1 j
X X X
b−1−b b−1 ∗
cj rj (s − µ + µrj ) = ci ri − A (s) gi cj rjn−1
j=1
s − µ + µr j i=1 i=1 j=1
b
X b
X b
X b
X b
X
cj rjb ∗
gi rjb−i rj−1 ci rib−1 ∗
cj rjn−1

− A (s) = − A (s) gi
j=1 i=1 i=1 i=1 j=1
b
X b
X b−1
X b
X
cj gi rjb−i−1 = gi cj rjb−1−i
j=1 i=1 i=1 j=1
b
X b
X
cj gi rjb−i−1 = 0
j=1 i=1
b
X
cj gb rjb−b−1 = 0
j=1
b
X cj
= 0 since gb ̸= 0
j=1
rj

Similarly putting n=b − 2, b − 3, ....., 1 in (16),we get the following set of b − 1

equations:
b b b b b
X cj X cj X cj X cj X cj
= 2
= 3
= ....... b−2
= =0
j=1
rj r
j=1 j
r
j=1 j
r
j=1 j
rb−1
j=1 j

So,we have
b n rb − A∗ (s) Pb gi rb−i o
j
X i=1 j
p∗n (s) = cj rjn−b , n≥1
j=1
s − µ + µrj

Setting s → 0,we get

b n rb − Pb gi rb−i o
j
X i=1 j
pn = p∗n (0) = cj rjn−b , n≥1
j=1
µ(rj − 1)

Now adding (18)-(20) we get,

∞
X ∞
X ∞
X ∞
X b−1 X
X n
s p∗n (s) = pn (0) + µ p∗n (s) −µ p∗n+1 (s) ∗
− A (s) gi pn−i (0)
n=0 n=0 n=1 n=0 n=1 i=1
∞ X
X b
∗
− A (s) gi pn−i (0)
n=b i=1
2.3 GI X /M/1 MODEL 24

∞
X ∞
X b−1 X
X n ∞ X
X b 
s p∗n (s) = ∗
pn (0) − A (s) gi pn−i (0) + gi pn−i (0)
n=0 n=0 n=1 i=1 n=b i=1
∞
X ∞
X b−1 X
X b−1 X∞ X b−1 ∞
X 
s p∗n (s) = ∗
pn (0) − A (s) gi pn−i (0) + gi pn−i (0) + gb pn−b (0)
n=0 n=0 i=1 n=i n=b i=1 n=b
∞
X ∞
X X ∞
b−1 X X∞ 
s p∗n (s) = pn (0) − A∗ (s) gi pn−i (0) + gb pn (0)
n=0 n=0 i=1 n=i n=0
∞
X ∞
X b−1
X X∞ ∞
X 
s p∗n (s) = ∗
pn (0) − A (s) gi pn (0) + gb pn (0)
n=0 n=0 i=1 n=0 n=0
∞
X ∞
X b
X ∞
X
s p∗n (s) = pn (0) − A∗ (s) gi pn (0)
n=0 n=0 i=1 n=0
X∞ ∞
 1 − A∗ (s)  X
p∗n (s) = p − n(0)
n=0
s n=0

Let s → 0 and using A∗ (0) = 1

Also define

pn = p∗n (0)

∞
X X
pn = A∗(1) (s) s = 0 n = 0∞ pn (0)
n=0

 1X ∞
1=− − pn (0)
λ n=0

∞
X
pn (0) = λ
n=0

Now we have

b
pn (0) 1X n
p−
n = P∞ = ci r
k=0 pk (0) λ i=1 i
2.3 GI X /M/1 MODEL 25

and

∞
X
p0 = 1 − pn
n=1
∞ X b n rb − Pb
X j gi rjb−i o n−b 
i=1
=1− cj rj
n=1 j=1
µ(rj − 1)
Xb n rb − Pb gi rb−i o X ∞
j i=1 j
=1− cj b
rjn
j=1
µ(rj − 1)rj n=1
b
X n r b ( P b −i
i=1 gi rj − 1)
o r
j j
=1− cj b
j=1
µ(1 − r )r
j j 1 − rj
b
X cj rj (G(rj−1 ) − 1)
p0 = 1 −
j=1
µ(1 − rj )2

Now

∞
X
Mean(L) = npn
n=1
∞ b n rb − Pb gi rb−i o
j
X X i=1 j
L= n cj rjn−b
n=1 j=1
µ(r j − 1)
b n rb − Pb gi rb−i o X ∞
j
X i=1 j
= cj b
nrjn
j=1
µ(rj − 1)rj n=1
b b −i
X n rjb ( i=1 gi rj − 1) o
P
rj
= cj b
j=1
µ(1 − rj )rj (1 − rj )2
b
X cj rj (G(rj−1 ) − 1)
L=
j=1
µ(1 − rj )3
2.3 GI X /M/1 MODEL 26

Also,
∞
X ∞
hX i2
2
Variance(V ) = n pn − npn
n=1 n=1
∞ b n rb − Pb gi rb−i o b
X X j i=1 j
X cj rj (G(rj−1 ) − 1) 2
= n 2
cj rjn−b −
n=1 j=1
µ(rj − 1) j=1
µ(1 − rj )3
b n rb (Pb gi r−i − 1) o r (1 + r ) X b
cj rj (G(rj−1 ) − 1) 2
j
X i=1 j j j
= cj −
j=1
µ(1 − rj )rjb (1 − rj )3 j=1
µ(1 − rj )3
b b
X cj rj (G(rj−1 ) − 1)  1 + rj  X cj rj (G(rj−1 ) − 1) 2
= −
j=1
µ(1 − rj )3 1 − rj j=1
µ(1 − rj )3
b b
X cj rj (G(rj−1 ) − 1)  1 + rj  X cj rj (G(rj−1 ) − 1) 
V = −
j=1
µ(1 − rj )3 1 − rj j=1
µ(1 − rj )3