Queuing Theory” The queuing theory, also called the waiting line theory, owes its development to A K Erlang’s efforts to analyse telephone traffic congestion with a view to satisfying the randomly arising demand for the services of the Copenhagen automatic telephone system, in the year 1909. The theory is used in situations where the ‘customers’ arrive at some ‘service station(s)’ for some service; wait (occasionally not); and then leave the system after getting the service. In such ‘arrival and departure” problems, the customers mi ht _be le waiting to deposit their electricity bills at a cash counter, machines waiting to be repaired in a factory’s repair shop, aeroplanes waiti i patients in a hospital who need treai --- and so on. The service stations in sud blems are the cash counters in thi Tepairmen in the shbp, runways at the airport and doctors attending the patients respactivaly, me more examples of queuing situations are given in Table 9.1, The waiting lines develop because the service to a customer may not be rendered immediately as the customer reaches the service facility. Thus, lack of adequate service facility would_cause waiting lines of customers td be formed. The only way that the service demand can be met with eas ie icity (and raising the efficiency of the existing iF possible) to the existing level. The capacity might be built to such high level as can always meet the peak demand with no queues. But adding to capacity may be a costly affair and uneconomit because then it shall remain idle to vai a manager, therefore, has to decide on an appropriate level of service which ie neither too Tow nor too high. Providing too low service would cause excessive Stig WHE Vas cat rcs of clube eoebor Tenders the Tong fun, divect cost of idle employees (where, for examples thee rpooeee have to wait near the store to obtain the supplies of materials, parts or tools needed for their work), or loss associated with yr employee morale resulting from being idle. On,the other hand, a c Wor in very Nigh set up cost and idle time for the service station(s). Thus, the goal ol ing, modelling Is the achievement of an economic balance between the cost of providing service and the cost associated with the wait required for that service.aioe TABLE 9.1 Queuing Examples Situation Arriving Service Customers Facility (a) Passage of customers Shoppers Checkout counters through a supermarket checkout (>) Flow of automobile Automobiles Road network traffic through a road network (© Transfer of electronic Electronic Transmission lines messages messages (@ Banking transactions Bank patrons Bank tellers (e) Flow of computer programmes = Computer Central processing through a computer programmes unit system (Q Sale of theatre tickets Theatregoers Ticket booking windows {g) Arrival of trucks to carry Trucks Loading crews and fruits and vegetables facilities from a central market (h) Registration of unemployed Unemployed Registration at employment exchange personnel assistants @ Occurrences of fires Fires Firemen and equipment @ — Flowofships tothe Ships Harbour and docking seashore facilities (K) Calls at police control Service calls Policemen room GENERAL STRUCTURE OF A QUEUING SYSTEM The general structure of a queuing system is depicted in Figure 9.1. Inpyt Population Service System Customers Leave Queue the System Structure Service Mechanism FIG. 9.1 General Strcture of the Queuing System We shall discuss in more details the various elements of a queuing symand then present mathematical results for some specific systems. The elements of a system are : Arrival Process The arrivals from the input population may be classified on different bases as follows : According to source The source of customers for a queuing system can be infinite or finite. For example, all people of a city or state a jal customers at a . The number of people being very large, it can be taken to be infinite. On the other hand, there are many situations in business and industrial conditions where we cannot consider the population to be infinite—it is finite. Thus, the ten machines in a factory uiring repairs and maintenar« ulation like r finite population. Removing one machine from a small, finite, population lik is_will have a noticeable effect on the calls expected ta be made (for this_will have a noticeable effe repairing) by the remaining machines than if there were a large number of by the remaining machines | SS machines, say 500. i ice C8) According to numbers The customers may_arrive for servic! individua i oups., Single arrivals are illustrated by customers visiting a_beautician, students reaching at a library counter, and so. on. On the other hand, families visiting restaurants, ships discharging cargo at a dock are examples of bulk. or batch, arrivals —, ‘c)) According to time Customers may arrive in the system at known (regular or otherwise) times, or they might arrive in a random way. The queuing models wherein customers" arrival times are known with certainty ape ategorized as deterministic_models (insofar as this characteristic’ is ei _——— concerned) and are easier to handle. On the other hand, a substantial majority of the queuing models are based on the premise that the customers enter the of the queuing models are based on the premiss system stochastically, at random points in time. With random arrivals, the number of customers reaching the system per unit time might be described by a probability distribution. Although the arrivals might follow any pattern, the frequently employed assumption, which adequately supports many real world situations, is that the arrivals are Poisson distributed. , Service System_ There are two aspects of a service system—{a) structure of ‘service system, and (b) the speed of service. (a) Structure of the service system By structure of the service system we mean how the service facilities exist. There are several possibilities. For example, there may be (i) A single service facility A library counter is an example of this. The models that involve a single service facility are called single server models. Figure 9.2 (a) illustrates such a model. Service Facilit Arrivals Queue Clstomers FIG. 9.2 (a) Single Server, Single Queue Model(ii) Multiple, parallel facilities with single queue That is, there are more than one server. The term parallel implies that each server provides the same type of facility. Booking at a service station that has several mechanics, each handling one vehicle, illustrates this type of model. It is shown in Figure 9.2 (b). Service Stations Customers Suave Leave Arrivals. FIG. 9.2 (b) Multiple, Parallel Servers, Single Queue Model (iii) Multiple, parallel facilities with multiple queues This type of model is different from the earlier one only in that each of the servers has a different queue. Different cash counters in an electricity office where the customers can make payment in respect of their electricity bills provide an example of this type of model. Figure 9.2 (c) portrays such a model. Arrivals___ Queues Service Stations Customers Leave ~— 000-4 —O000+—{_ +} — OOO0C—{_ FIG. 9.2 (c) Multiple Servers in Series (iv) Service facilities in a series In this, a customer enters the first station and gets a portion of service and then moves on to the next station, gets some service and then again moves on to the next station ... and so on, and finally leaves the system, having received the complete service. For example, machining of a certain steel item may consist of cutting, turning, knurling, drilling, grinding, and packaging operations, each of which is performed by a single server in a series. Figure 9.2 (d) shows such a situation. Service Station Service Station Queve [Customers Arrivals__Queve FIG. 9.2 (4) Multiple, Parallel Servers, Multiple Queues ModelBesides these, there may be other possibilities as well. ; - | (b) Speed of service Tn a queuing system, the speed with which sevice is provided can be expressed in either of two ways—as service rate an ) service time. The service rate describes the number of customers serviced during a particular time period. The service time indicates the amount of time needed to service a customer. Service rates and times are reciprocals of each other and either of them is sufficient to indicate the capacity of the facility. Thus, if a cashier can attend, on the average, to 10 customers in an hour, the service rate would be expressed as 10 customers/hour and service time would be equal to 6 minutes/customer. Generally, however, we consider the service time only. If these service times are known exactly, the problem can be handled easily. But, as generally happens, if these are different and not known with certainty, we have to consider the distribution of the service times in order to analyse the queuing system. Generally, the queuing models are based on the assumption that service times are exponentially distributed about some average service time. 3. Queue Structure Another element of a structure. In the queue structure, discipline which means the order queuing system is the queue the important thing to know is the queue by which customers are picked up from the waiting line for service. There are a number of possibilities. They are: (a) First-come-first-served When the order of service of customers is in the order of their arrival, the queue discipline is of the first-come-first: type. For example, board the bus first. (b) Last-come-first-served _ Sometimes, order reverse of the order in which they enter so that the ones who join the last are served first. For example, assume that letters to be typed, or order forms to be processed accumulate in a pile, each new addition being put on the top of them. The typist or the clerk might process these letters or orders by taking each new task from the top of the pile. Thus a just arriving task would be the next to be serviced provided that no fresh task arrives before it is Picked up. Similarly, the people who join an elevator last are the first ones to leave it. (c) Service-in-random-order (SIRO) Random order of service is defined as: whenever a customer is chosen for service, the selection is made in a way that every customer in the queue is equally likely to be selected. The time of arrival of the customers is, therefore, of no consequence in such acase. (d) Priority service The customers in a queue might be rendered service on a priority basis. Thus, customers may be called according to some identifiable characteristic (length of served with a queue at the bus stop, the People who came first will the customers are serviced in an customers may be classi customers join a queue, 8 (a) patient, or (b) impatient. If the ‘ when it exis Station for Betting servic, its, and wait till they enter the service €, they are called patient customers. On the otherhand, the queuing systems may enjoy customer behaviour in the form of defections from the queue. The customers may not select queues randomly (if there are multiple queues) and look for the shortest queue. There may be jockeying among the many queues, that is the customers may switch to other queues which are moving ‘fast’, and also reneging is possible—when a customer stands in the queue for some time and then leaves the system because it is working ‘too slowly’. There may also be bribing or cheating by some customers for queue positions. Besides, some customers may, upon their arrival, not join the queue for some reason and decide to return for service at a later time, or may even abandon the input population altogether. In terms of the queuing theory, this is known as balking, and occurs particularly when there are limits on the time and the extent of storage capacity available to hold waiting customers. Unless otherwise specified, the storage capacity is taken to be infinite. In the queuing models that we consider, we shall assume that there is no balking or jockeying and that the customers leave the system only after receiving service, and not before. Mathematical models give way to simulation when this assumption breaks. OPERATING CHARACTERISTICS OF A QUEUING SYSTEM An analysis of a given queuing system involves a study of its different operating characteristics. This is done using queuing models. Some of the more commonly considered characteristics are discussed below. 1. Queue length—the average number of customers in the queue waiting to get service. Large queues may indicate poor server performance while small queues may imply too much server capacity. 2. System length—the average number of customers in the system, those waiting to be and those being serviced. Large values of this statistic imply congestion and possible customer dissatisfaction and a potential need for greater service capacity. 3. Waiting time in the queue —the average time that a customer has to wait in the queue to get service. Long waiting times are directly related to customer dissatisfaction and potential loss of future revenues, while very small waiting times may indicate too much service capacity. 4. Total time in the system —the average time that a customer spends in the system, from entry in the queue to completion of service. Large values of this statistic are indicative of the need to make adjustment in the capacity. 5. Server idle time —the relative frequency with which the service system is idle. Idle time is directly related to cost. However, reducing idle time may have adverse effects on the other characteristics mentioned above. We now proceed to discuss some of the queuing models. It may be mentioned here that the results obtained from various models are based on the assumption that the service system is operating under equilibrium or steady state conditions. For many systems, the operating day begins in transient state with stomers in the system. It takes some initial time interval for enough nO cuss to arrive such that a steady state balance is reached. It should be customeinderstood that a steady state does not mean that the system will ed point where the number of customers in the system never changes. Evenwhen fe system reaches equilibrium, fluctuations will occur. A steady state condition really implies that various system performance measures (the operating characteristics) would reach stable values. THE MODELS The queuing models can be categorised as being deterministic or probabilistic If each customer arrives at known intervals and the service time is known’with certainty, the queuing model shall be deterministic in nature. The vast majority of the queuing models are, however, based on the assumption that one or more elements of the queuing system can be expressed only in probabilistic terms. Hence, nearly all of the queuing models are of probabilistic type. _ an: seal Ae Deterministic Queuing Modely.~ Vipul. visue = Let us first consider the case where the customers arrive in the queuing system at regular intervals and the service time for each customer is known and constant. visual he Suppose that customers come to a bank's teller counter every 5 minutes. Thus the interval between the arrival of any two successive customers is exactly 5 minutes. Suppose further that the banker takes exactly 5 minutes to serve a customer. Here the arrival and the service rates are each equal to 12 customers per hour. In this situation there shall never be a queue and the banker shall always be busy with work. Now, suppose that the banker can serve 15 customers per hour. The consequence of this higher service rate would be that the banker would be busy __» 4/5th of the time and idle in 1/5th of his time. He shall take 4 minutes tow serve a customer and wait for 1 minute for the next customer to come. There would be, as before, no queue. visual If, on the other hand, the banker can serve only 10 customers per hour, then the result would be that he would be always busy and the queue length will increase continuously without limit with the passage of time. It is easy to visualize that when the service rate is less than the arrival rate, the service facility cannot cope with all the arrivals and eventually the system leads to an explosive situation. The problem in such situations can be resolved by providing additional service station(s). Symbolically, let the arrival rate be A. customers per unit time and the service rate is p customers per unit time. Then, if App the waiting line shall be formed which will increase indefinitely; the service facility would always be busy; and the service system will eventually fail; and if A

1, ~~ the system would ultimately fail, and . . if ps1, the system works and p is the proportion of time it is busy.We can easily visualize that the condition of uniform arrival and uniform service rates has a very limited practicability. Such conditions May exist when we are dealing, for example, with movements of items for Processing in highly automated plants. However, generally, and more Particularly when human beings are involved, the arrivals and servicing time are variable and uncertain. Thus variable arrival rates and servicing times are the more realistic assumptions. The probabilistic queuing models, as mentioned Previously, are based on these assumptions. Probabilistic Queuing Models Of the numerous queuing models available, we shall consider the following models: (a) Poisson - exponential, single server model - infinite population; (b) Poisson - exponential, single server model - finite population; and (c) Poisson - exponential, multiple server model - infinite population. In each of these, the words ‘Poisson-exponential’ indicate that the customer arrivals follow Poisson distribution while the service times are distributed exponentially. These theoretical distributions are discussed in chapter 2. To recapitulate, if the arrivals are independent, with the average arrival rate equal to 4 per period of time, then, according to the Poisson probability distribution, the probability that n customers will arrive in the system during a given interval T, is given by the following: P (n customers during period T) = e-™ 4% ne where m= AT, ande=2.7183 = —— - Cc—e—— 1, Poisson-Exponential Single Server Model-Infinite Population This model is based on the following assumptions: (a) The arrivals follow Poisson distribution, with a mean arrival rate of, say, A. (b)_ The service time has exponential distribution. Let the average service rate be p. (<) Arrivals are from infinite population. (d) The customers are served on a first-come-first-served basis. (e) There is only a single service station. To evaluate the model, we begin with the first question whether the service station can handle the customer demand for service. The answer to this question lies in the values of j1 and A . If 4.2, the waiting line would increase without limit, leading to the breakdown of the system ultimately. For a workable system, it is necessary that 1 < p. It is interesting to observe that even if } = 1, the system shall not work. This can be shown mathematically but 1s beyond the scope of this book. As pointed out earlier, the ratio p = A/u is defined as the traffic intensity. This indicates the proportion of time, or the probability, that the service station is busy. From this, the probability that the system is s idle, that is, there are no Customers in the system, equals Pp = 1p. We can show that the probability of Raving exactly one customer in the system is P, = pPo. Similarly, the Probability of having exactly 2 customers in the system would be P, = p P; = p”Pp. To generalize, the probability of having exactly n customers in the system shall be: 7 P= pes php) We can use this result to determine the expected number of customers in the system, an een aan 0.4 = ynp,This can be solved to obtain P Vv aA or 4a or 1-p Expected number of customers in the queue shall be equal to the difference between the expected number of customers in the system and the expected number of customers being served. Now, since the server is busy, or is serving one unit, A/p of the time, and is serving 0 unit (i.e., it is idle) 1-A/p of the time, the expected numbering served equa (3) oof iad. my Consequently, expected number of s in the queue, =—_4 9 w(u-A) -p tt L, is the avera: It is significant to note that thy is the average length of all queues inchuding empty queues. The average léngth of non-empty queues, that is, those which contain at least one customer, L,’ is give allowing expression. ~ With an average arrival rate of 1>the gE time between the arrivals is 1/2. Thus the mean waiting time in queue, W,, is the product of the average time between the arrivals and the average queue length. Symbolically, 1 Sr 7 Substituting 27/y(u - A) for L, and simplifying we get =k 2 W, = a= —o. ap tk Similarly, the mean time in the system, Wy, is equal to the H i Product of the average ee between arrivals and the average number of customers in the WaLxLyF Putting L, = 4,/( - ) in this equation and simplifying, we get w= x pe Stated another way, since the_mean_ servic, is (expected) time for completing the sery 1 Therese a ae io ne a customer would spend in the system shall be equal to the expected waiting time in the queue plus Processing/servicing time. Thus,jo 4 =—+_,1__1 rt nu-2) HO pHa a which is the same as shown earlier. Finally, the probability that a customer spends more than ¢ units of time in the system, W, (#) =e°/™", And, the probability tha a customer speaie esse than t units of time i W,O=pe™ ti ——_ It is slants 46 acts Wei cienerece tem Spent in the system is distributed exponentially, and so is the service time, but the difference of the two, that is to say, the time spent in the queue, W, (t), is not exponentially distributed. ; Mas ne fk cle See er aaN 21 Poisson-Exponential Single-Server Model - Finite Population:— This queuing model is based on similar assumptions as the earlier discussed model except that the input population is finite. For this model, the system structure is such that we have a total of M customers; a customer is either in the system (consisting of a queue and a single service station), or outside the system and, in a sense, arriving. In particular, when a customer is in the arriving condition, then the time it takes him to arrive is a random variable having an exponential distribution with mean = 1/A. As a result, when there are n customers in the system, then there are M - n customers in the arriving state and, therefore, the total average rate of arrivals in the system is } (M - n). It may be observed that this system is self-regulating. By this, we mean that when the system gets busy, with many of these customers in the queue, then the rate at which additional customers arrive is, in fact, reduced, thus, lowering the further congestion in the system. %To illustrate, suppose that a factory has five machines which are required to be handled by the maintenance staff for repairing and working. The number of customers (the machines here), or the population, is finite, a small countable set. It has the implication that the removal of one machine from production can drastically change the probability of another machine being removed. Thus, there exists a strong dependency relationship between the arrivals and they are no longer independent. Of course, if there were a large number of machines in the factory, say three hundred, removal of one machine would have but an insignificant impact on the chances of another machine being removed for repairs. In such a situation, the assumption of independence of arrivals nearly sustains and, therefore, we can use the first model for studying various operating characteristics of the queuing system. For the reason of dependency relationships between arrivals, the Poisson probability law cannot be strictly applied when the input population is finite. Instead of the arrivals’ statement as an average for the population, we classify them as an average of a unit time. Hence, as mentioned, the inter- arrival time of customers is stated to be following exponential distribution with mean = 1/1.LOTT Te te ee REIL FYE PEUUULE @ LUDL SAVILLE, U1 £270. - _Poisson-Exponential Multiple Server Model - Infinite Population Each of e previous two models were based on the assumption that there is only one service facility in the system. Now we shall consider a model in which there are multiple service stations, which are giving parallel, identical service. It covers situations where, for example, there may be more than one doctor in a hospital whom the patients can visit; there may be more than one runway at an airport for takeoff and landing, ... and so on. In cases where multiple servers are involved, an important point to consider is whether there is a separate, independent queue for each server or the customers form a single queue from which they are picked up for service. Indeed if there are as many queues as the service stations, equal to, say, K, the system can be thought of, and analysed as one composed of K different single server queuing systems. For our purposes here, we shall presume that the customers form a single queue. Assumptions The model is based on the following assumptions: (a) The arrival of customers follows Poisson probability law, the average arrival rate being A, (b) The service time has an exponential distribution, . (c) There are K service stations, each of which provides identical service,