QUEUEING THEORY Introduction Whenever customers line up at a service facility, a queue system exists.

Queues are common place experiences. Some examples that can easily be recognized are: People waiting in shops and at service counters; Cars waiting at traffic lights; Aircraft waiting to land;

Students waiting to see an academic adviser;

Ships waiting to enter port. The less obvious examples of queues are: Machines waiting for repair; Goods waiting in inventory; Telephone subscribers waiting for a clear line;

Papers in an in-tray.

A branch of management science that concerns itself with the general phenomenon of customers lining up for service is queue theory. A theory of queues is possible because certain regularities about the patterns of arrival and service of customers exist. These regularities can be described statistically and analytical results (i.e. formulae) sometimes follow. Queuing theory is about delay whether or not an actual queue is observed.

Why does delay occur at all? The reason for delay is that it is not possible or not worthwhile to tailor the supply of a service exactly to the demand for it. A trade-off situation arises. While delay can be costly, it is also costly to make provision to avoid or to reduce delay. However, these two costs do not always fall on the same parties. Sometimes a system of serving customers can be developed by experimentation. For example, changing the provision of cashiers at supermarket checkouts at different times of the day or banking hall at different periods of the month is possible. However, in some other cases, such intervention in the real situation would either be impossible or prohibitively expensive. This is particularly so where large or irreversible investments are involved, for example, the number of lanes on a motorway cannot readily be changed. Thus, very complex systems have to be simulated. In principle, queue theory can predict how systems will operate. Certain important measures of performance (or system parameters) can be obtained from queue models. Among these are: average waiting time expected length of queue average number of customers in the system probability of experiencing delay. The three main means of obtaining these parameters are by:

 Analytical results Real world experimentation Simulation In practice, queue theory in pure analytical form is employed relatively infrequently. However, where it is used, it tends to be used extensively and it has often been found to be of powerful effect. Characteristics of Queueing System Queueing systems (waiting lines) are characterized by the differing nature of five factors: 1.
2.

The arrival pattern of customers The queue discipline The service mechanism Capacitation Population

3. 4.
5.

Arrivals The pattern of arrivals may be deterministic (e.g. items on a production flow line) or more usually random (e.g. vehicles entering filling station, customers at the post-office, supermarket or bank). Customers may arrive one at a time or en-masse. Customer or item are the general terms used to describe the queue elements as they do not have to be people. Arrivals may or may not depend on the state of

the system (e.g., a customer turning back on seeing a long queue). The arrival rate may be constant or may vary over time. Customers may be identical or different (e.g., civilians and military men arriving at a bank, private and distressed aircraft arriving at an airport).

Queue Discipline The simplest arrangement of queue discipline is FIFO (first in, first out). There is also LIFO (last in, first out, as in items drawn from stock or redundancies in work force) and random (as with telephone callers trying to get a connection). In the case where there are several queues, customers may wish to change from one queue to another (jockey). Also, some customers may join a queue and then leave before service (renege). These two cases are elements or signs of queue indiscipline. Service Mechanism There may be one or many servers, who may differ in speed of service. The speed of service at any service point may be constant or random and may vary with the time of the day. Servers may be in parallel (as in a self-service cafeteria) or in series. Capacitation

Capacitation refers to the capacity or maximum available space in the queueing system. Some systems have a maximum number of customers that can be contained in the system. Where such a limit exists the system is said to be CAPACITATED. The theory of queues only takes cognizance of customers within the queue system and not those waiting outside the system. Population Population refers to the source from which customers come into the queue system. The population from which customers arrive may be infinite or finite. The population is sometimes referred to as the calling source. While a very large number of different queueing models can be constructed, those of interest and of practical relevant to us have patterns of arrival and service described by probability distributions rather than being deterministic. A purely random pattern of arrivals is described by the Poisson Distribution and its continuous analogue the negative exponential distribution. The Poisson Distribution In any time interval T, the probability that there are n arrivals into the queueing system is given by:

( T ) n e T P (n arrivals in time interval T) = n!

Probability that there are exactly n customers in the system (steady state).

Average number of customers arriving into the system per unit time. Average number of customers a server can service per unit time. The traffic intensity or proportion of busy time of the server. L The expected (or average) number of customers in the system. Lq The expected (or average) number of customers waiting in the Queue. W The expected time a customer will spend in the system (both waiting on the queue and being served). Wq The expected time a customer will spend waiting in the queue. A single Channel Queue Model

The simplest queueing model is referred to as the single channel exponential queueing model and it is written in shorthand notation as the (M/M/1) system. The assumptions of this model are as follows: 1.The input population is infinite (that is, there are an infinite number of potential customers (arrivals). 2.The arrival (or input) distribution is Poisson distributed with average arrival rate . This means that the number of customers (arrivals) arriving in one unit of time is a random variable that has a Poisson probability density. The average arrival rate per unit time is customers. This is equivalent to saying that the time between successive customers arriving is a random variable having an exponential probability density with mean 1/ . In the notation (M/M/1), the first M means that the arrival pattern is Poisson distributed.. 3.The service discipline is first come, first served (FCFS).

4.The service facility consists of a single channel. The one (1) in (M/M/1) means one channel (or one server). 5.The service distribution is also Poisson. The mean number of customers served per unit of time by a busy server is . Equivalently, we can say that the service time per customer is a random variable having an exponential probability density with a mean of
. In the
1

notation (M/M/1), the second M means that the service rate is Poisson. 6.The average number of customers arriving (arrivals) per unit of time is assumed to be less than the average number of customers who could be served by the system per unit time, that is, < . 7.System capacity is infinite. 8.Balking and reneging are not allowed. Steady state Equations for the (M/M/1) System The following equations can be shown to hold for the (M/M/1) queue system:
8

Traffic density, = (1) Pn = n (1- ) system. The expected (or average) number of customers in the system, L, is given by: L= (3) The expected (or average) number of customers waiting in the queue, Lq is: Lq = ( ) = [2/(1 )] (4) The expected time a customer will spend in the system (both waiting and being served), W, is given by: W (5) The expected time a customer will spend waiting in the queue, Wq, is =
1

(2)

where Pn is the probability of there being n customers in the

= [/(1 )]

Wq (6) Traffic Intensity

( )

The traffic intensity is denoted by and is given by equation (1). Traffic intensity is the proportion of time that the server will be busy serving customers who arrive. Sometimes, it is referred to as the utilization factor of the queueing system. Since < , then < 1. Idle time For what proportion of the time will the server be idle? If equals the busy time, then 1 is the idle time. This can be seen from Eq. (2). The server is idle when there are no customers in the system. The probability of there being no customers in the system is o, which from Eq.(2) yields: 0 = o(1- ) = 1(1- ) = 1 - .

Mean Service Time If the average number of customers served per unit time is , then the mean (or average) service time per customer is
1

time units. For example, if = 10 customer per hour,

10

then mean service time is =

1 1 0

= 0.10 hour (per

customer) or 6 minutes per customer. Example 1 Joes service station operates a single gas pump. Cars arrive according to Poisson distribution at an average rate of fifteen cars per hour. Joe can service cars at the rate of twenty cars per hour on the average, with service time per car following an exponential probability distribution. Assuming we model Joes service station as an (M/M/1) queueing system, compute the steady-state statistics for the system. Solution We have = 15 cars per hour, and Thus, 1) Traffic intensity, = =

15 20

20 cars per hour.

= 0.75

This implies that Joe will be busy servicing cars 75 percent of the time. This leaves an idle time of 1- = 0.25, or 25 percent, for Joe to do paperwork, maintenance, and so on. 2) Expected (average) number of cars is the system, L. L=

15 20 15

= 3 cars

The implication of this is that on the average Joe can find three cars at his station.
11

3)

Expected (or average) number of customers waiting in the queue: Lq = ( ) =

= 2.25

So on the average 2.25 cars will be waiting in line for service. The practical interpretation of this is that while two cars are still on queue, it remains of the tank of the car currently receiving service to be filled. Expected time a customer will spend in the system (i.e. waiting and being served): W=
1 1 = 20 15

4)

= 0.20 hour

Or 12 minutes
5)

Expected time a customer will spend waiting in the queue, Wq is: Wq = ( ) =

= 0.15 hour

Both (4) and (5), respectively, imply that on the average, a car spends 0.20 hour in the system (or 12 minutes) and 0.15 hour waiting for service (9 minutes).

12

Suppose we wish to find the probability of having four cars or less in the system, this probability equals the sum of the various probabilities from zero cars being in the system to four cars being in the system, that is 0 + P1 + P2 + P3 + P4. Note that from Equation (2): Pn = n(1- ) = (0.75)n (1-0.75) = (0.75)n (0.25) Thus, we have ----------------------------------------------n Pn = (0.75)n (0.25) (0.75)0(0.25) = 0.250 (0.75)1(0.25) = 0.188 (0.75)2 (0.25) = 0.141 (0.75)3(0.25) = 0.105 (0.75)4 (0.25) = 0.079 = 0.763 --------------------------------------------The above result is interpreted to mean there is a 76.3 percent chance of finding four or fewer cars line up at the service station. ---------------------------------------------0

1 2 3 4

13

The above example can be represented pictorially as shown below:

Departing Arriving cars

Car

Service facility: a single gas pump serving 20 cars per hour

Key relationships of the steady-state statistics In every Poisson queueing system, certain relationship will always hold between L, Lq, W, and Wq. These are as follows: L = W Lq = Wq W = Wq +
1

(7) (8) (9)

Each of the above relationships holds for the (M/M/1) system. Example 2 Customers arrive in a shop at the average rate of 42 per hour. The shopkeeper takes on average one minute to serve each customer. Simple queue conditions thus prevail.

14

(i)

What proportion of his time does the shopkeeper spend serving customers? What is the average queueing time for customers? What is the probability that:

(ii) (iii)

a)There are three customers in the shop? b)There is one customer queueing? c)There are four or more customers in the ship? d)There are three or less customers in the shop? Solution i)This refers to the proportion of the time when there is at least one customers in the system. This is given by 1 + P2 + P3 + .. + P . More conveniently, this is the same as 1- P0. In fact it is itself, the traffic intensity or utilization factor. Thus,

42 60

= = = 0.7

1minute to serve

Since it takes 1 minute to serve each customer, then the number of customers served in 1 hour is 60. (ii) Average queueing time is the same as average time a customer will spend in the queue, Wq. Wq = ( ) =

42 50 (50 42 )

15

7 180

hours or

1 3

minutes

(iii)(a) We shall assume here that being in the shop means being in the queue system. Then the probability of three (3) customers being in the shop is 3 = 3(1- ) = (0.7)3 (1-0.7) = (0.7)3 (0.3) = 0.1029 (b) If one customer is queueing this must mean that one is being served so that there are two customers in the system. Thus, 2 = 2(1- ) = (0.7)2(1-0.7) = (0.7)2 (0.3) = 0.147 (c ) Here we have simply: P (4 or more) = 4 = (0.7)4 = 0.2401 (d ) To simplify workings we note that three or less and four or more cover all possibilities. Thus, (3 or less) = 1- (4 or more) = 1-0.2401 = 0.7599 Example 3
16

A telephone is used by eight people per hour on average and the mean call length is three minutes. A further instrument has been requested but this will only be installed if average queueing time is at least three minutes. Determine the percentage increase in demand necessary to produce this result. Solution We note that = 8 callers per hour. Clearly, the service time is the length of call so that with an average call length of three minutes, there will be 20 callers per hour. Thus = 20 callers per hour. Expected queueing time is given by ( ) , thus Queueing time = ( ) = (1) If average queueing time is 3 minutes = hours (2)

1 20 3 60

20 (20 )

hours =

1 20

Thus, we require the R.H.S of (1) and (2) to be equal. Hence = 20 (20 )

This (by cross multiplication) implies that 20 = 20 (20 - ) or That is, 20 by transferring = 400 - 20
the 20 on the R.H.S to the 20 + 20L.H.S 400 =

17

40 = 400

= 10 callers per hour

Increase in callers per hour Initial average arrivals = 25 percent.

Thus, a 25 percent increase in demand (measured by rate of arrivals) would be needed to install a further instrument.

18