Queueing Theory

ALISH VIJI VARGHESE ASSISTANT PROFESSOR OF MATHEMATICS COLLEGE OF ENGINEERING THALASSERY November 24, 2011

Contents
1 Queueing Theory 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Syllabus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Denition . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.3 Application of Queueing Theory . . . . . . . . . . . . . . 1.1.4 Elements of Queueing System . . . . . . . . . . . . . . . . 1.1.5 Operating Characteristics of Queueing System . . . . . . 1.1.6 Distributions in queuing systems . . . . . . . . . . . . . . 1.1.7 Classications of Queueing Models . . . . . . . . . . . . . 1.2 Poisson Queueing Systems . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Model I- (M/M/1) : (/F IF O) . . . . . . . . . . . . . . 1.2.2 Littles Formula . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Model 2 (M/M/1) :( N/FIFO) . . . . . . . . . . . . . . . 1.2.4 Model 3 (Generalized Model: Birth-Death Process) . . . . 1.2.5 Model 4 (M/M/C) :( /FIFO) . . . . . . . . . . . . . . . 1.2.6 Model 5 (M/M/C) :( N/FIFO) . . . . . . . . . . . . . . . 1.2.7 Model 6 (M/M/C): (C/FIFO) . . . . . . . . . . . . . . . 1.2.8 Model 7 (M/M/R) :( K/GD, KR)-Machine-Repairman model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Non-Poisson Queueing Models . . . . . . . . . . . . . . . . . . . . 1.3.1 Model 1 (M/G/1): (/GD) . . . . . . . . . . . . . . . . . 1.4 Problums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 5 5 5 5 6 7 7 8 9 9 9 10 10 11 12 12 12 13 13 14

CONTENTS

Chapter 1

Queueing Theory
1.1
1.1.1

Introduction
Syllabus

Basic structure of queueing models, exponential and Poisson distribution, the birth and death process , queueing models based on Poisson input and exponential services time, the basic model with constant arrival rate and service rate, nite queue, limited source Q models involving non exponential distributions, single service model with Poisson arrival and any services time distribution, Poisson arrival with constant service time, Poisson arrival with constant service time, Poisson arrival and Erlang service time priority disciplines.

1.1.2

Denition

The basic phenomenon of queueing arises whenever a shared facility needs to be accessed for service by a large number of jobs or customers. : Bose

1.1.3

Application of Queueing Theory

Telecommunications, Trac control, determining the sequence of computer operations, Predicting computer performance, Health services (eg. control of hospital bed assignments), airport trac, airline ticket sales etc(At retail stores and banks, a single queue is more ecient than separate queues before each server.)

CHAPTER 1. QUEUEING THEORY

Figure 1.1: Queueing System

1.1.4

Elements of Queueing System

Input or Arrival Process (a)Size of the queue No of potential customers waiting to be served. It can be nite or innite .arrival may occur in batches of xed or variable size or one by one. (b)Arrival distribution This gives the pattern by which the arrivals take place. It is usually described by the arrival rate or inter-arrival time. These are characterized by probability distributions. Commonly arrival rate follows Poisson distribution or inter-arrival time follows exponential distribution. (c)Customers behavior if a customer decides not to enter the queue because of the queue size, then he is said to be balked. On the other hand if a customer decides to move from the queue after entering it because of impatience, he is said to have reneged. If a customer moves from one queue to another for personal gains, he is jockey for positions. Input process can bestationary or transient (time dependent). Queue disciplines It is the rule by which customers are selected to service when a queue is formed. The most common queue discipline is rst come rst out (FCFO) or rst in rst out (FIFO).Last in rst out (LIFO) and service in random order (SIRO) are the other queue disciplines. In priority discipline the service given according to some priority schemes. The service in priority disciplines are of two types:pre-emptive (customer with higher priority is served rst.) and non-pre-emptive (customer with lower priority is served rst.).In case of parallel channels fastest server rule (FSR) is adopted.

1.1. INTRODUCTION Service Mechanism

If there is innite number of servers then all customers will be served on arrival and there will be no queue. If number of customers is nite, then customers are served in some specic order. They may serve in batches of xed or variable size called bulk service system. The service system is described in terms of service rate. Service facilities can be of the following types 1. Single queue one server, 2. Single queue several servers, 3. Several queues one server, 4. Several servers arranged in parallel channels (barber shop with more than one chair) or series channels. Capacity of the system. The source from which costumers are generated may be nite or innite. The size of the queue also can be nite or innite.

1.1.5

Operating Characteristics of Queueing System

The following are some important characteristics of a queueing system of general interest. 1. E(n) orL or Ls :- Expected number of customers in the system.(both in queue and service) 2. E(m) or Lq :- Expected number of customers in the queue.(excluding that in service) 3. E(v) or W or Ws :- Expected waiting time in the system.(service time and waiting time in queue) 4. E(w) or Wq : - Expected waiting time in queue till the beginning of service 5. P : - Server utilization factor or busy period or trac intensity. It is the proportion of time the server spends with the customers which is given by / where is the arrival rate and is the service rate or departure rate.

1.1.6

Distributions in queuing systems

Poisson Distribution The following assumptions are made in a Poisson process. Axiom 1 : The number of arrivals in non-overlapping intervals is independent of what happened prior in time. Axiom 2 : The probability of more than one arrival in a small time interval t is negligible Axiom 3 : The probability that an arrival occur between time t and time t + t is equal to t + O(t)

CHAPTER 1. QUEUEING THEORY

Distribution of arrivals (pure birth models): Let Pn (t) denote the probability of n arrivals in a time interval of length t. If the above axioms are satised, then Pn (t) is given by Pn (t) = (t)n t e f or n 0 n! (1.1)

This is the Poisson distribution with mean t

Distribution of inter-arrival time Inter arrival time is the time between two successive arrivals. If arrival rate follows the exponential distribution then the inter arrival time follows the exponential distribution f (t) = et . The mean inter arrival time is 1/.

Distribution of departure As in the case of arrival rate, if Pn (t) is the probability that n customers remaining after t times. If is the service rate or departure rate and N is the number of customers at time t=0 then

Pn (t) =

(t)N n t e :1nN (N n)!

(1.2) (1.3)

P0 (t) = 1
n=1

Pn (t)

This is known as truncated Poisson Law

1.1.7

Classications of Queueing Models

Any queueing system can be completely specied by the symbolic form: (a/b/c) : (c/e). First and second symbols (a&b) stands for type of distributions of interarrival times and inter-service times. Third symbol species the number of servers. Forth symbol stands for the capacity of the system and the last for the queue discipline. The following symbols are used to specify some important distributions. M Poisson arrival or departure distribution. EK Erlangian or gamma inter-arrival or service time distribution. GI General input distribution. G General service time distribution.

1.2. POISSON QUEUEING SYSTEMS

1.2
1.2.1

Poisson Queueing Systems

This is single channel, Poisson input, exponential service time with innite system capacity. This type of queueing models can be solved in three steps. The same procedure can be applied to other models also. Step I: Construction of dierential-dierence equations. Step II: Deriving the steady state dierence equations. Step III: Solution of the steady-state dierence equations. The nal step will gives the steady state solution as Pn = Where = length.

= (1 )

(1.4)

< 1 and n > 0. This gives the probability distribution of queue

Characteristics of Model 1 Probability of queue size being greater than n is given by n . Average number of customers in the system is E(n) = Ls = Average queue length is given by E(m) = Lq =
2 1 1

2 () () , andP (m

Average length of non empty queue E(m/m > 0) = 0) = P (n > 1) = (f rac)2 The variance of queue length is given by V (n) =
(1)2

>

()2

Average waiting time of a customer in queue is given by E(w) = Wq = (1) = () Average waiting time of a customer in the system including service time 1 E(v) = Ws = () Average waiting time of a customer who has to wait E(w/w > 0) =
1 ()

1.2.2

Littles Formula

These formulae give relationships between average queue length and average waiting time. They are; Ls = Ws (1.5) Lq = Ws and Ws = W q + 1 (1.6) (1.7)

10

CHAPTER 1. QUEUEING THEORY

Note: The above mentioned characteristics are unchanged expect for service time distributions even if the queue discipline is SIRO or LIFO. So we write GD to represent these three disciplines.

1.2.3

Model 2 (M/M/1) :( N/FIFO)

In this model the maximum number of customers in the system is limited to N. as in the case of model 1, by solving the dierence dierential equation in steady state we have the following result Pn = (1 )n = 1; 0 n N 1 N +1 1 = =1 N =1

(1.8)

[These steady state probabilities exist even for 0 ] Characteristics of model 2 Average number of customers in the system Ls = [1 (N + 1)N + N n+1 ] (1 )(1 N +1 )

Average number of customers in the queue Lq = 2 [1 N n1 + (N 1)n ] (1 )(1 N +1 )

The waiting time in the system and queue can be calculated using Littles 1 formula, i.e. Ws = Ls where = e = (1 PN ) and Wq = Ws

1.2.4

Model 3 (Generalized Model: Birth-Death Process)

This model deals with a queueing system having single service channel, Poisson input with no limit on the system capacity. Let n= number of customers in the system n = Arrival rate of customers given n customers in the system. n =Departure rate of customers given n costumers in the system. Pn =Steady-state probability of n customers in the system. Under steady state conditions for, n 0, the rates of ow into and out of state n must be equal. From the transition-rate diagram the balance equation is given by. n1 P n 1 + n + 1Pn+1 = n Pn + n Pn n1 1 P1 = 0 P 0 n=0

1.2. POISSON QUEUEING SYSTEMS and solving them we have the steady state probability as Pn = n1 n2 0 P0 n n1 1 n1
n=0

11

and the value of Pn can be obtained by using the condition Special Cases

Pn = 1

Case I When n = for n 0; and n = for n > 1 then the case become model 1. Case II When n = for n 0 and n = n for n > 1 1 P0 = e and Pn = n! n e for n > 0 This model is known as Self-service model and is represented by (M/M/) : (/F IF O).For this model Ls = and Lq = 0; Wq = 0.

1.2.5

Model 4 (M/M/C) :( /FIFO)

Here we have C parallel service channels with service rate per service channel, arrival rate is . In eect the service rate of the service faculty is n, if n < C and C if n C. using generalized model , the steady state probabilities are given by Pn = 1 n P0 n! 1 = nC n P0 C C! 1nC (1.9) n>C

Characteristics of model 4 Average queue length is given by Lq = (/)C P0 (C 1)!(C )2

Average number of customers in the system is given by Ls = Lq +

Expected waiting time and average time customers spend in the system can be found using Littles formula Probability that an arrival ha to wait P (n C) = (/)C CP0 C!(C )

Probability that an arrival enters service without wait 1 P (n C)

12

CHAPTER 1. QUEUEING THEORY

1.2.6

Model 5 (M/M/C) :( N/FIFO)

In this model the maximum number of customers in the system is limited to N (N C). The steady state probabilities are given by Pn = P0 1 = nC C C! 1 n!
n

0n<C
n

(1.10) P0 CnN

Characteristics of model 5 Average queue length is given by Lq = P0 (C)C 1 N C+1 (1 )(N C + 1)N C C!(1 )2

Average number of customers in the system is given by

Ls = Lq + C P0
n=0

(C n)(C)n n!

Average waiting time in the system and queue can be obtained by Littles formula E(v) = [E(n)]/ E(w) = E(v)
1

= (1 PN ) or E(w) = f racE(m)

1.2.7

Model 6 (M/M/C): (C/FIFO)

In this model number of service channels and system size are same. Thus no queue is allowedin this model The stationary distribution here is known as Erlangs rst formula and can be obtained from above by putting N = C. thus Pn = 1 (/)n P0 n!

(1.11)

= 0 otherwise

1.2.8

Model 7 (M/M/R) :( K/GD, KR)-Machine-Repairman model

Here we have R identical service channels.The maximum number of incoming customers is limited to K. In other words the population from which arrivals occur is of nite size.

1.3. NON-POISSON QUEUEING MODELS Characteristics of model 7 Average number of customers in the system is given by
R1

13

Ls = P0
n=0

K n

1 R!

n
n=R

K n! n RnR R1

Average queue length

Lq = Ls R +
n=0

(R n)

K n

Littles formula can be used to nd the waiting time in system and queue with = e = [K E(n)]

1.3

Non-Poisson Queueing Models

In such models arrivals or departures or both may not follows Poisson axioms. Analysis of such models is more complicated because the Poisson axioms do not hold. The following techniques are used to analysis a non-Poisson queue. i Phase technique:-it is used when the arrival demands phases of services ii Imbedded Markov chain techniques:-here non-Markovian queues are reduced to markovian queue. iii Supplementary Variable technique:-this is a technique to convert to Markovian types. The models analyses using these techniques include GI/G/C, M/G/I, GI/M/S, GI/Ek/I, D/Ek/I, M/Ek/I, M/G/I.

1.3.1

Model 1 (M/G/1): (/GD)

This model consists of a single server, Poisson arrival and general service time distribution. The characteristics of this model are as given below Average number of customers in the system Ls = the variance of the service time distribution. Average queue length Lq =
2 2 +2 2(1) . 2 2 +2 2(1) 2 2 +2 2(1)

+ where 2 is

Average waiting time of a customer in the queue Wq = Average waiting time in the system Ws =
2 2 +2 2(1)

14

CHAPTER 1. QUEUEING THEORY

1.4

Problums

1.1 A branch oce of a large engineering rm has one on-line terminal that is connected to a central computer system during the normal eight-hour working day. Engineers, who work throughout the city, drive to the branch oce to use the terminal to make routine calculations. Statistics collected over a period of time indicate that the arrival pattern of people at the branch oce to use the terminal has a Poisson (random) distribution, with a mean of 10 people coming to use the terminal each day. The distribution of time spent by an engineer at a terminal is exponential, with a mean of 30 minutes. The branch oce receives complains from the sta about the terminal service. It is reported that individuals often wait over an hour to use the terminal and it rarely takes less than an hour and a half in the oce to complete a few calculations. The manager is puzzled because the statistics show that the terminal is in use only 5 hours out of 8, on the average. This level of utilization would not seem to justify the acquisition of another terminal. What insight can queueing theory provide? 1.2 A mechanic looks after 8 automatic machines; a machine breaks down, independently of others, in accordance with a Poisson process. The average length of time for which a machine remains in working order is 12 hours. The duration of time a machine required for repair has an exponential distribution with mean 1 hour. Analysis the situation using queueing theory. 1.3 Discuss non Poisson queueing models