Queuing Models‘The queuing theory or waiting line theory owes its development to A.K. Erlang. He, in 1903, ‘ook up the problem on congestion of telephone traffic. The difficulty was that during busy periods, telephone operators were unable to handle the calls the moment they were made, resulting, in delayed calls. ‘Waiting lines or queues are omnipresent. Businesses of all types, industries, schools, hospitals, cafeterias, book stores, libraries, banks, post offices, petrol pumps, theatres — all have queuing problems. Queues are also found in industry — in shops where machines wait to be repaired, in ‘Waiting line problems arise either because @ there is too much demand on the facilities so that we say that there is an excess of waiting time or inadequate number of service facilities (ii) there is too less demand, in which case there is too much idle facility time or too many facilities. In cither case, the problem is to cither schedule arrivals or provide proper mumber of facilities or both so as to obtain an optimum balance between the costs associated with waiting time and idle time,Customers z “aniving * |_____ Queue or c waiting ine Service facility }+———. System ———_——_> Customers leaving ig Cotcioeey as suenore oe u a8 Waiting time cost i Increased Service —» 3. 10.5 Total cost of operating service faci.ELEMENTS OF A QUEUING SYSTEM (STRUCTURE OF A QUEUING SYSTEM) A queuing system is specified completely by seven main elements: Input or arrival (inter-arrival) distribution Output or departure (service) distribution Service channels Service discipline Maximum number of customers allowed in the system Calling source or population Customer's behaviour. aw1. Arrival Distribution. It represents the pattern in which the number of customers arrive at the service facility, Arrivals may also be represented by the inter-arrival time, which is the period Detween two successive arrivals Arrivals may be separated by ual intervals of Gime oF by unedual Dat definitely known intervals of time or by unequal intervals of time whose probabilities are known; these are called tandom arrivals. The rate at which customers arrive to be serviced, je., number of customers arriving per unit of time is called arzival rate, When the arrival rate is random, the customers arrive in no logical pattern or order over time. This represents most cases in the business world. ‘When arrivals are random, we have to know the probability distribution describing arrivals, specifically the time between arrivals, Management scientists have demonstrated that random rivals are often best described by the Poisson distribution ee Se ee ‘Mean value of arrival rate js represented by A, Itmay be noted that the Poisson distribution with mean arrival rate 2 is equivalent to the (negative) exponential distribution of inter-arrival times with mean inter-arrival time 1s,2. Service (Departure) Distribution. It represents the pattern in which the number of customers leave the service facility. Departures may also he represented by the service (inter~ departure) time, which is the time period between two successive services. — Service time may be constant or variable but known or random (variable with only known probability) If service times are randomly distributed, we have to find out what probability distribution best describes their behaviour. Jn manv cases where service times are random. management scientists hhave found that they are best described By the exponential probability distribution. ‘The rate at which one service channel can perform the service, i.e., number of customers served per unit of time is called service rare. This rate assumes the service channel to be always busy. i.e., no idle time is allowed. Mean value of service rate is represented by u. In business3. Service Channels. The queuing system may have a single service channel. Arriving customers may form one line and get serviced, as in a doctor’s clinic. The system may have a number of service channels, which may be arranged in parailel or in series or a complex combination of both. In case of parallel channels, several customers may be serviced simultaneously, as in_a barber shop, For series channels, a customer must pass successively through all the channels before service is completed, e.g, a product undergoing different processes over different machines or students, during admissions going through one counter after another before all admission formalities are complete. A queuing model is called one server model, when the system has one server only and a multi-server model when the system has a number of parallel channels each with one server.4. Service Discipline, Service discipline or order of service is the rule by which customers are selected irom the queue for service. The most common discipline is ‘first come, first served’, according to which the customers are served in the order of their arrival, eg., cinema ticket windows, railway stations, banks, ete. The other discipline is ‘last come, first served’, as in a big godown, where the items arriving last are taken out first. Still other disciplines mélude ‘service in random order (SIRO)’ and ‘priority’. ‘Priority’ is said to occur when an arriving customer is chosen for service ahead of some other customers already in the qucvc. = —eReSSe5, Maximum Number of Customers allowed in the System (Capacity of the System). Maximum number of customers in the system can be either finite or infinite, In some facilities, aly a limited number of customers are allowed in the svstem and new arr not allowed to join the system unless the number becomes less than the limiting value, 6. Calling Source or Population. The arrival pattern of the customers depends upon the source which generates them. If there are only a few potential customers. the calling source (population) js called finite, If there are a large number of potential customers (say, over 40 or 50), itis ull id to be infinite. There is still another rule for categorising the source as finite or infinte-A finite soure exists when an arrival affects the probability of arrival of potential future customers. For example, a battery of M running machines is a finite source, as far as machine repair situation is concemed. Before any machine breaks down, the calling source consists of M potential customers. As soon as a machine breaks down, it becomes a customer and hence cannot generate another ‘call’ until it gets serviced (repaired). An infinite source is said to exist when the arrival of a customer does not affect the rate of arrival of potential future customers.7. Customer's behaviour: The customer's behaviour is also very important in the study of queues. If a customer decides not to enter the queue since itis too long, he is said to have balked. If a customer enters the queue, but after sometime loses patience and leaves it, he is said to have —Leteged_Whgn there are two o more parallel queues and the eusiomers move from one quve to the other, they are said to be jockeying.OPERATING CHARACTERISTICS OF A QUEUING SYSTEM Analysis of a queuing system involves a study of its different operating characteristics Some of them are 1. _Quewe length (Iq) ~ the average number of customers in the queue waiting to get service. This excludes the customer(s) being served 2, System length (L,) ~ the average number of customers in the system including those waiting as well as those being served. 3. Wanting time in the queue (WV,) ~ the average time for which a customer has to wait in the queue to get service 4. Total time in the system (W,) ~ the average total time spent by a customer in the system from the moment he arrives till he leaves the system. It is taken to be the waiting time plus the service time i 5. Utilization factor (p) ~ it is the proportion customers. It is also called traffic intensity f time a server actually spends with theKENDALL’S NOTATION FOR REPRESENTING QUEUING MODELS DG Kendall (1953) and later A. Lee (1966) introduced wseTaT notation for queuing models ‘The complete notation can be expressed as (albto)- (del). rival Gr ntorarival diseaBUTOR, — departure (or service time) distribution, ‘number of parallel service channels inthe system, service diseipine, ‘maximum number of customers allowed in the system, calling source or population ‘The following conventional codes are generally used to replace the symbols 6 and d- Symbols for a and b = ign (Poisson) arrival or departure distribution (or exponential interarival or seRvT6ET ition), E,©Erlangian o gamma interarrival or service time distribution with parameter & = general independent arrival distribution, general departure distribution, ‘D = deterministic interasval or seviee times, Symbols for d CES = frst come, frst served, LCS = Tost come, SIRO = service in random order, GD= general sevice discipline os.¢ and frep customers in the system The symbols ¢ and frepresen finite Q) or infinite (=) numberof customers inthe system, and calling source respectively, For instance, (M/E,/1) : (FCES/Nic) represents Poisson arival (exponential interarvaly, Erlangian departure, single server, ‘frst come, frst served” discipline, maximum allowable customers N in the system and infinite population model where KaasCLASSIFICATION OF QUEUING MODELS {a) Probabilistic Queuing Models — 2 Moe! 1 viene Model) This model is sumbolicaly present by (M/M/1) Tae). This Fepresens Poisson arrival (exponential interarrval), Poisson departure (exponential service time), single server, fist come, first served service discipline, infinite tuumber of customers allowed in the system and infinite population, Since the Poisson and exponential distributions are related to each other, both of them are denoted by the symbol "MF due to Markovian property of exponential distribution 2. Model If (General Erlang Model) : Though this model is also represented by (M/M/1) (FCFS/eo/e0), it is a general queuing model in which the arrival and service rates depend ‘upon the length of the queue. Some persons desiring service may not join the queue since itis too long, thus allecting the arival rate. Sunilary, service rate is also allected by the Tength of the queue. 3. Model IIT: This model is represented by (M/M/1) : (SIRO/co/ee) It is essentially the same ‘as model I except that the service discipline is SIRO insetad of FCFS. 4, Model IV : This model is represented by (M/M/I) :(FCFS/Néo). in this model the capacity of the system is Timited of file say N. So the number of srivals cannot exceed N. Model V : This model is represented by (MAW/1) : (FCFSiniM). Its finite-popalation or limited source model. In this model the probability of an arrival depends upon the number of potential customers available to enter the system. 6 Model VI: This model is represented by (/Mle) : (FCFS/so/se). This is same ax model except that ere ae ¢ service channels working in parallel. MSIE are ¢ service ch(b) Deterministic Model 1, Model XI: This model is represented by (D/D/1) : (FCFSf/ee). In this model interartival time as well as service time are fixed and known with certainty. The model is, therefore, called deterministic model, (©) Mixed Queuing Model 12, Model XII : This model is represented by (M/D/1) : (FCFS/v). Here, arrival rate is Poisson distributed while the service rate is deterministic or constant.SINGLE-CHANNEL QUEUING THEORY A single-channel queuing problem results from random interarrival time and random service time at a single service station. The random arrival time can be described mathematically by a probability distribution. The most common distribution found in queuing problems is Poisson distribution. This is used in single-channel queuing problems for random arrivals where the service 10.9-1 Models for Arrival and Service Times Generally, arrivals do not occur at fixed regular intervals of times but tend to be clustered or scattered in some fashion. A Poisson distribution is a discrete probability distribution which redicts the number of arriv: jiven time. The Poisson distribution involves the probability ‘of occurrence of an arrival. Poisson assumption is quite restrictive in some cases. It assumes that arrivals are random and independent of all other operating conditions. ‘The mean arrival rate (Le., the number of arrivals per unit of time) A is assumed to be constant over time and is independent of the num er OF units already serviced, queue length or any other random property of the queue.Since the mean arrival rate is constant over time, it follows that the probability of an arrival between time and 1+ di is. de Thus probability of an arrival in time dt = % dt 10.1) The following characteristics of Poisson distribution are written here without proof Gaye Probability of m arrivals in time ¢= n= 0,12, 0.2) nl Probability density function of inter-arrival time (time interval between two consecutive arrivals) =A em, (10.3) Finally, Poisson distribution assumes that the time period dl is very small so that (di), (dé), ete. — 0 and can be ignored.Service time is the time required for completion of a service j,g., it is the time interval bet beginning of a service and ils completion. The mean service rate is the number of cust d per unit of ime (assuming lhe service to bE GORITIGUE Trougiiout the entire ime unit, while the average service time I/ is the time required to serve one customer. The most common ty of distribution used for service times is exponential distribution, It involves the probability of ‘completion of a service. It should be noted that Poisson distribution cannot be applied to servicing because of the possibility of the service facility remaining idle for some time. Poisson distribution tssumos fixed fe interval of continuo servicing, which ean never be assed inal sree. ‘Mean service rate yt is also assumed to be constant over time and independent of number of units already serviced, queue length or any other random property of the system. Thus probability that a service is completed between ¢ and ¢ + df, provided that the service is continuous = ud. Under the condition of continuous service, the following characteristics of exponential distribution are written, without proof qu” .e Probability of m complete services in time ¢ = = (0.4) Probability density function (p.d,f) of interservice time, ic., time between two consecutive services =p. ¢ ™. (10.5)10,9-2 Model |. Single-Channel Poisson Arrivals with Exponential Service, Infinite Population Model [(MIMM) : (FCFS/=/=)]_ Let us consider a single-channel system with Poisson arrivals and exponential service time distribution, Both the arrivals and service rates are independent of the number of customers in the waiting linc. Arrivals are handled on “first come, first served" basis. Also the mean errival rate A is less than the mean service rate wt. ‘The following mathematical notation (symbols) will be used in connection with queuing, models: = number of customers in the system (waiting line + service facility) at time & = mean amival rate (number of arrivals per unit of time). = mean service rate per busy server (number of customers served per unit of time) ‘dt= probability that an arrival enters the system between f and f+ df time interval fe, within time interval di | = Aate= probability that no arrival enters the system within interval dt plus higher order terms in di. .ean service rate per channel, Hadr= probability of one service completion between ¢ and £ + dt time interval i. within time interval dt 1 -wadt= probability of no service rendered during the interval dr plus higher order terms in dt—_Pn= steady state probability of exactly » customers in the system p,{t) = transient state probability of exactly nm customers in the system at time ¢, assuming the system started its operation at time zero Pru = transient state probability of having m+ | customers in the system at time 1. Pys(@) = transient state probability of having m — 1 customers in the system at time 1. p,At + dt) = probability of having n customers in the system at time t+ dt L, = expected (average) number of customers in the queue. a SPEIER Keven ef) Bieber Nesom 1. = expected number of customers in the system (waiting + being served). W, = expected wailing time per customer in the queue (expected time a customer eps waiting in line). W, = expected time a customer spends in the system. (in waiting + being served) L,= expected number of customers waiting in line excluding those times when the line is empty i.e., expected number in non-empty queue (expected number of customers in a queue that is formed from time 10 time). W,,= expected time a customer waits in line if he has to wait at all i-e., expected “time in the queue for non-empty queueTo determine the properties of the single-channel system, it is necessary to find an expression for the probability of m customers in the system at time f ie., p,(0), for, if p,(@) is known, the expected number of customers in the system and hence the other characteristics can be calculated. In place of finding an expression for p,(), we shall first find the expression for p,(t + di) ‘The probability of mr units (customers) in the system at time t + dt can be determined by summing up probabilities of all the ways this event could occur. The event can oceur in four mutually exclusive and exhaustive way's TABLE 10.1 ‘Event No. of units at |No. of arrivals in] No. of services in | No. of units at time time dt time dt time t+ dt L n 0 0 n 2 net 0 1 n 3 nl 1 0 ” 4 n 1 1 nNow we compute the probability of occurrence of each of the events, remembering that the probability of a service or arrival is ude or Adt and (dt)? — 0. . Probability of event 1 = Probability of having m units at time ¢ x Probability of no arrivals x Probability of no services = pat)(1 ~ det) (1 — de) = py(t) (1 — dt — wd + w (dt?) = pal) [1 = Adi ~ we). Similarly, probability of event 2 = p,.1(0). (1 — Ade) . (ude) = Pr(® [adr], probability of event 3 = Pri Dat]. — nde) = pps [Ade], probability of event 4 = prt). (A dt) (u. dt) = pat) [ew (ab?| = 0.Note that other events are not possible because of the small value of dt that causes (dt)? to approach zero (as in event 4), Since one and only one ofthe above events can happen, we can obtain p(t + dt) [where n > 0} by adding the probabilities of above four events. 2 pa* dd) = pd) [I~ Balt — wal] + Pye) (Hd + Py [at] + 0 or pad + dt) = ppl) — palO [Ade + Wt] + Presi) [edt] + py sO [ee] ox BLED BD 21). p46 pelt LO Taking the limit when di — 0, we get the following differential equation which gives the relationship between Py Py-1,Pyst at any time f, mean arrival rate and mean service rate d Sp PADI= APD + Wine * WPA, where n> 0.107)‘After solving for p,(t + dt) where n > 0 itis necessary to solve for py(t + dt) where n= 0 se. to solve for po t+ db IP n= 0, enly two mutually exclusive and exhaustive events can occur as shown in table 10.2 TABLE 10.2 Foon No.of units at [No.of arrivals n] No. af services in] No. of wns at dime t time dt time de tine t+ dt 1 0 0 e 0 2 L o 1 ° Probability Probability Probability Probability ofevent | = of having of moamivals of no serviees no unit at = pf) * (l= Add) «1 and Probability Probability Probabil Probability ofevent2 = of havingone <= af noamivals of one service = plo) * (= 2d = (u dd, [Note that if no units were inthe system. the probability of no service would be 1. Probability of having no unit in the system at time « + dt is given by summing up the probabilities of above two events, ult + dt) = put) (1 = Aa) + py(t) (ud) (1 — Aa) rt) ~ pol.) * pf). (at) old Cat) + py(0) (ua) Pil — Apu or Pult + di) — rfl bolt + dd)~ pelt) a‘When dt» 0, te differential equation which indicates the relationship between probabilities ‘po and p, at any time f, mean arrival rate 2 and mean service rate is d 5p f= mes) ~ Aes), where m = 0 (10.8) Equations (10.7) and (10.8) provide relationships imolving the probability density fnetion ‘pdt forall values of n but sill we do not know the value of pal) Assuming the steady state condition for the system, when the probability of having units (eustomers) in the system becomes independent of time, we get d PAO = Pm ae AO] = 0.‘Therefore, for a steady y state system the differential equations (10.7) and (10.8) reduce to difference equations (10.9) id (10.10) Abe + Wee ~ G+ WP, where n> 0, (109) py -Apo where n =O. (10.10) # Putting m= 1 in equation (10.9), we have O= Apo + up2— (+ Hp) Similarly, for m = 2, equation (10.9) gives -( 4 Ps ( a Po Pn ie) Po Where n> 0. ao.10.12) (10.13)Having known the value of p,, we ean find the various operating characteristics ofthe system, L. Expected number of units in the system (waiting + being served), L, is obtained by using the eck ot ‘expected value: ‘The series within brackets is an infinite series ofthe form 0. a, 2a, 3a, ., xa. For such an infinite series, if a and less than one, the sum is given by the formula Pe eae (a) se (-Dietel: oth 10.14)2. Expected number of units in the queue, 1-y~ Expected number of units in the system — spect none ae gle Sa in service (single server -2= [oe | H nna Note that the expected number in service is | times the probability that the service channel (10.18) # busy te, 1. % is busy te, 1. ~ “ 3._Expected time per unit in the system (expected time a unit spends in the system). Expected mumber of units in the system _ Ly a 1 016) “Arrival rate RG a nH 4. Expected waiting time per unit in the queue, W, = Expected time in system — time in (10.17)ree] (070 hems 8) bent “CDE O- ‘Variance of queue length 10.18)6. Average length of non-empty queue (length of queue that is formed from time to time), Ly For a on-enply queue, he miner cunts inthe system should be a least 2 (one inservice and the other in the queue). Probability of a non-emy [Now average length of non-empty queue, Average length of queue 2 Probability of non-empty queue 7 waiting time in non-empty queue (average waiting time of an arrival who waits), or expected waiting lime per busy period. 10.20) 10.19)8, Probability of quewe be in or equal tok Wot pi tert + Peal a a9.__ Probability of queue being greater than k, row=(4) 021b) 10, Probability tha the quewe 16 non-empss Pin>)=1~ py io2te) 11. Probability density function of wating time (excluding service) distribution Kune 10 =| 10.22) zu» Lt H 12. Probability density function of (waiting + service) time distribution FMW. eee 0.23)A self-service store employs one cashier at its counter. Nine customers arrive on an average every 5 minutes while the cashier can serve 10 customers in 5 minutes. Assuming Poisson distribution for arrival rate and exponential distribution for service time, find 1. Average number of customers in the system 2. Average number of customers in the queue or average queue length. 3. Average time a customer spends in the system. 4. Average time a customer waits before being served. [.N.T-U. Hyderabad B.Tech. (Mech.) May, 2012; May, 2011; PIU. BE., 2001; Karn. U. B.E. (Mech.) 1998, 95]Solution Arrival rate A = 9/5 = 1.8 eustomersiminute, service rate = 10/5 = 2 customers/minute, 1. Average number of customers in the system, 18 Ley te 2. Average number of customers in the queue, we a A 18 18 em eG 2 2-188 3. Average time a customer spends in the system, 4, Average time a costumer waits in the queue, ACA 18(_1_) a ‘GH “2 (azia) ~ 45 minutesArrival rate of telephone calls at a telephone booth are according to Poisson distribution, with an average time of 9 minutes between two consecutive arrivals. The length of telephone call is assumed to be exponentially distributed, with mean 3 minutes. (a) Determine the probability that a person arriving at the booth will have to wait. (b) Find the average queue length that is formed from time to time. (©) The telephone company will install a second booth when convinced that an arrival would expect to have to wait at least four minutes for the phone. Find the increase in flow rate of arrivals which will justify a second booth, What is the probability that an arrival will have to wait for more than 10 minutes before the phone is free ? (e) What is the probability that he will have to wait for more than 10 minutes before the phone is available and the call is also complete ? (d) (D Find the fraction of a day that the phone will be in use.Solution Here, arrival ate, = 19 per minute, service rate. = 13 per minute (a) Probability that person will have to Aug 1 AMO ogy win 3 7° (8) Average queve lent tha is formed fom time to time ws 3 19 wm 173-179" 2793" 2 ~ 1S Persons. (©) Let be the new (increased) arrival rte to justify the installation ofthe second telephone booth Average waiting time in the queue = ——"1—— ame ape Gh) T03-—) or or hy «Tt vs. 12 =F srivatsininwe 79 Increase in low rate of arrivals = = Sper minute ce 29S(a) Probability [waiting ime 2 10) = fe (1-2). at (©) Probbility ime in sytem 210] = 70-2) 2-9 a tore aoa ea . a [eo wa ppeeonp ed ne" 201 () The expeste fraction of day that the phone will be in useCustomers arrive at the First Class Ticket Counter of a Theatre at the rate of 12 per hour There is one clerk serving the customers at the rate of 30 per hour (What is the probability that there is no customer at the counter (i.e., that the system is idle)? (i) What is the probability that there are more than 2 customers at the counter ? (ii) What is the probability that there is no customer waiting to be served ? (iv) What is the probability that a customer is being served and no body is waiting ? Solution Here, = 12hour, u = 30/hour, (® Probability that there is no customer in the system, 2 Po ~ 706. (i) Probability that there are more than tsvo customers at the counter + Pat pst ~ Bo + Pi + Pa)[roa Pan [ose 2.5] «nie (ui) Probability that there ie no customer waiting to be served * probability that thee is at most one customer at the counter (») Probability that a customer is being served and no body is waiting es “ rnp 2 os] 24