1 CHAPTER-1 BASICS AND HISTORICAL DEVELOPMENT OF QUEUING THEORY (This chapter deals with the
fundamental concepts of queuing theory, Economic interpretation of queue, literature survey,
motivation along with probability distribution functions, fuzzy set theory with fuzzy numbers and scope of the thesis). 1.1 INTRODUCTION: The history of mankind shows that the way we live our lives has been largely driven by great innovations. Man has always formed novel ways of meeting all upcoming requirements whether it is advent of the wheel or the personal computer or information technology devices. The way these, revolutionizing tools have been used brought a profound impact on the human society. The thirst for knowledge and improving the quality of life has led us to a better understanding of the environment we live in. Numerous reasons caused man to colonize a venture that would have been unsuccessful if he had not learnt tolerance, especially the patience of waiting for others. One of them stopped somewhere and waited for others to join. It was this tolerance and patience of waiting for others because of which the colonies came into existence. With the growth of the world‟s population man‟s colonies grew into urban communities of various sizes. He started leading an ordered life as he had to rely on the services of others to a large extent to satisfy his own demands .If man had followed a philosophy of „immediate service on demand‟, the result would have been an uneconomical utilization of his total efforts, interpersonal harmony would never have got developed and his problems would have been numerous. Waiting, therefore, became a part of life and waiting line thus is common place usually formed by units, customers, messages that queue 2 up awaiting service while the presence of waiting lines is very obvious and predictable. The prediction about the detailed behavior can be made possible only by a systematic study of waiting lines. 1.2 BASICS OF QUEUE: Queues (waiting lines) are a part of everyday life. We all wait in queues to buy a movie ticket, make a bank deposit, pay for groceries, mail a package, and obtain food in a cafeteria, vehicles waiting at petrol pumps or to pass a road bottle neck etc. A queue, in fact doesn‟t contain discrete units only, liquids entering in a sudden large volume into a tank having only a small outlet pipe can be said to be queuing in the tank. Customers wait when the number of customers requiring service exceeds the number of service facilities. An amazing number of waiting time situations exists as when computer programs are waiting to be processed at computer center. The queuing theory is the probabilistic study of waiting lines. Although it does not solve all type of waiting line problems, yet it provides useful and vital information by forecasting or predicting to various characteristics and parameters of the particular waiting lines under study. Since the prediction about the waiting times, the number of waiting at any times, the time for which the servers remain busy, delay etc. heavily depends on the basic concept of stochastic processes, hence, queuing theory can be taken as an application of stochastic processes. Queuing theory is generally considered a branch of Operational Research because the results are often used when making decisions about the resources needed to provide services. Thus we can say queuing theory is a stochastic process in Operations Research. The queues and queuing systems have been the subject of considerable research since the appearance of the first telephone systems. 3 1.3 ECONOMICAL ASPECTS OF QUEUING THEORY: Waiting for service is usually an unpleasant experience, which creates unhappiness among customers who may switch to competitive service offerings. Having to wait is not just a petty personal annoyance. Waiting time represents the loss of valuable resources and this loss translates into psychological as well as economical costs of waiting. The longer, customers wait the more dissatisfied they are to be. The amount of time that a nation‟s populaces waste by waiting in queues is a major factor in both, the quality of life and the efficiency of the nation‟s economy. Taking machines wait to be repaired may result in loss of production. Airlines waiting to take off or land may disrupt later travel schedule. Delaying service jobs beyond their due date may result in a lot of future business causing manufacturing jobs to wait to be performed may disrupt subsequent production. Queuing problems in industry are more serious than those experienced in our daily life. Excessive waiting time in production will mean longer manufacturing cycle, that is longer deliveries and high investment in work in progress. Virtually, all queuing situations have economic interpretation for their implication. In general there are two opposing economic aspects to queues. (a) It can cost money for an item to be idle in queue. An employee has to be paid while queuing at the tool store, valuable production is lost while machine stand out of order, tax receipts are delayed while document remain in the inspectors tray. And, of course, the smaller the service facility, longer the queues and higher the cost. (b) On the other hand, larger the service facility quicker it will disperse queue and therefore, more often the system will stand idle and hence it costs money to increase the service facility. 4 Service cost Waiting cost Number of Servers(s) Figure-1.1 Queuing theory based on probability concept gives an indication of the capability of a given system and of the possible changes in its performance with modification to the system. With the help of queuing theory the executive can at most make an informed guess of what could be the balance between the customer‟s waiting time and in preventing waiting time. He first takes into consideration several alternatives, evaluates through queuing models, their effects on the system and makes his choice. The criteria for evaluation will be the efficiency of the system like the average length of the queue, expected waiting time of a customer and the average time spent by the customer in the system 1.4 OBJECTIVE OF STUDY: The Queuing theory studies waiting line system by formulating the appropriate mathematical models of the particular system under study. These models and their operation are then used to derive the necessary performance measurements of the queuing system as a whole. These measurements provide the vital information which will be later utilized in designing an effective waiting line system. An appropriate balance or tradeoff between the various costs and the waiting for the required service are, of course, the optimum line in a waiting-line system. The total 5 cost of service and customer waiting time is the essence of our study of queuing systems. The optimum balance may be arrived at either by: (a) Introducing more facilities at extra cost, or (b) Changing the pattern of the arrival of customers for service, or Rescheduling arrival of customers (c) Changing pattern of servicing, or effecting method improvements to reduce service time. (d) To relate waiting delays queue lengths and other measure of performance to the given properties of the input stream and the service procedures. (e) To search out space cost along with customer waiting and idle service cost to obtain the optimal system design. (f) Ultimate objectives are to determine the increase in investment of service for a corresponding decrease in customer‟s delay. (g) To develop the queue model in order to predict fluctuating demands and to enable one to provide adequate service for the customers with tolerable waiting. (h) To search such queue models with control and delayed vacation so that the vacation period of servers can be utilized. (i) To minimize the total operations cost of the system in feedback queue model under parametric constraints. Thus the main reason for the study of queues is to enable us, through our queue models, to predict what will happen if certain changes are made. (e.g. in the pattern of arrivals, reducing the mean length of service time, increasing the number of 6 server or increasing the amount of room to accommodate waiting customers). Typical measures of congestion that we would like to be able to predict are the average number of customers waiting, the average length of time that they will wait, and the proportion of time that the service facility is busy. 1.5 ELEMENTS OF QUEUING SYSTEM: The basic feature of a queue system is as follows: 1. The Input or arrival-pattern of customers or units (λ-mean arrival rate). 2. The Output or service-pattern of customers or units (µ-mean departure rate). 3. The number of service channels. 4. The capacity of the system. 5. The traffic-intensity (Ratio of mean arrival rate to mean departure rate i.e. ρ=λ/µ) 6. Queue discipline (FIFO, LIFO, and SIRO etc.) Customer served customer The basic queuing process Figure 1.2 1.5.1 THE INPUT OR ARRIVAL-PATTERN OF CUSTOMERS: The input pattern means the manner in which the arrivals occur. It is specified by the inter arrival time between any two consecutive arrivals. A measure usually considered is the average length of the inter arrival time or its reciprocal, the Queue Service mechanism Input source 7 average number of arrivals per some unit of time. The input pattern also indicates whether the arrivals occur singly or in groups or batches. If in batches, the manner in which these batches are constituted is also to be covered in the same between any two consecutive arrivals, or it may be stochastic, when its distribution is also to be specified. Sometime an arrival may not that is, machines coming for repair whenever they fail. There may also be several classes of customers with different arrival rates. 1.5.1(a) CUSTOMER’S BEHAVIOR: Earlier researchers did not find significant place for impatient customers with Random service but customer‟s behavior play very important role in study of queues. If a customer decides not to enter the queue since it is too long he is said to be balked. If a customer enters the queue and leaves it, he is said to have reneged. When there are two or more parallel queues and the customers move from one queue to other, they are said to be Jockeying. 1.5.2 THE OUTPUT OR PATTERN OF SERVICE: By the pattern of service, we mean the manner in which the service is rendered. It is specified by the time taken to complete a service. The time may be constant (deterministic) or it may be stochastic. If it is stochastic, the pattern specification involves the distribution of service time of a unit. A measure typically considered is provided by the average time required to serve a unit or by the average no of units served per some unit of time. Sometimes service may be rendered in bulks or batches, as in the case of an elevator, instead of personalized service of one at a time. In this case, the manner of formation of batches for service also has to be specified. 8 1.5.3 THE NUMBER OF SERVERS: A system may have a single server or a number of parallel servers. An arrival who finds more than one free server may choose at random any one of them for receiving service. If he finds the entire server busy, he joins a queue common to all the servers. The first customer from the common queue goes to the server who becomes free first. This kind of situation is common- for example, in a bank or at a ticket counter. There may also be situations where there is a separate queue in front of each service facility, as in the case of a supermarket. There also arise cases of ordered entry when an arrival has to try to find a free server in the order the servers are arranged. 1.5.4 THE CAPACITY OF THE SYSTEM: A system may have an infinite capacity – that is, the queue in front of the server(s) may grow to any length. Against this there may be limitation of space, so that when the space is filled to capacity, an arrival will not be able to join the system and will be lost to the system. The system is called a delay system or a loss system, according to whether the capacity is finite or infinite. If finite, it will have to be specified by the number of places available for the queue as well as for the one(s) being served, if any. 1.5.5 TRAFFIC INTENSITY: The demand for service is often expressed as ρ=λ/µ, the mean number of arrivals per unit time taken as the mean service time. This quantity is called offered load (or traffic intensity); it is a dimensionless quantity and is expressed in Erlang. This offered load is a measure of what the customers want. It is also called utilization factor or server utilization. 9 1.5.6 QUEUE DISCIPLINE: Although most queues are based on a first come first serve (FIFO) system. This is not always so, sometimes certain queue components have priorities (e.g. ambulances at traffic lights), sometimes a LIFO system work (as when a tax inspector always takes the top document first from his in tray), and sometime components are served at random (SIRO) (as when a number of people persist in trying to get through to the same telephone number and successful caller is the one who is lucky enough to dial just at the moment the receiver is replaced after the previous call). There are, therefore different method of determining the order of service. The method appertaining to any situation is known as queue discipline. 1.6 TRANSIENT AND STEADY- STATE BEHAVIOR: A time-dependent solution of the stochastic behavior of queue process Pn(t), n≥0 is called transient state. It is often difficult to obtain such solutions. Or even when found, these may be too complicated to handle. For many practical situations, however, one needs the equilibrium behavior-that is, the behavior when the system reaches an equilibrium state after being in operation for a sufficiently long time. In other words, one is often interested in the limiting behavior of Pn(t) as t→∞. Denote Pn = lim p (t) n t , n= 0, 1, 2…… Whenever the limit exists, the system is said to reach a steady (equilibrium) state, and Pn is called the steady-state probability or the long-run proportion of time that the system contains exactly n customers. In steady state, the operating queue characteristics are independent of time. 1.7 TYPES OF UNCERTAINTITY: In general there are two types of uncertainties 10 1.7.1 PROBABILISTIC UNCERTAINTY: - Mostly whenever an event occurs, the uncertainty about its outcome is attributed to “randomness” or stochastic. The best to approach this question might be to describe the status in terms of probability distribution. Following processes are based on Probabilistic uncertainty. 1.7.1(i) STOCHASTIC PROCESS: A stochastic process is defined to be the process which develops in time according to some probabilistic law. The state space of a stochastic process is a list of possible outcomes at any point in time. We can say a stochastic process has a state space and a time structure. A stochastic process {X(t), t T} is a collection of random variables, that is, for each t T, X(t) is a random variable. The random variables X(t) denotes the state of the system at a time t. If t assumes discrete values, then the process {X(t), t = 0,1,2 … ……} is called a discrete parametric stochastic process. If t assumes continuous values then the process {X(t),t ≥0} is called a continuous parametric stochastic process. The set of possible values of X(t) is called state space and the set of possible values of t, denoted by T, is called parametric space or the index set. More clearly, a stochastic process is a family of random variables which provides description of the evolution of some physical phenomenon through time. Queuing systems provide many examples of stochastic processes. For example X(t) might be the number of customers that arrive before a service counter by time t, then {X(t), t≥0} is of type continuous time and discrete state space. Again it might be the queuing time of nth arrival, then {Wn; n= 0, 1, 2……..} is of type discrete time and continuous state space. Generally, the stochastic processes encountered in the analysis of a specified complex system can be identified any one of the following:- 11 (a) MARKOV PROCESS A stochastic (or random) system is called a Markov process if the occurrence of a future state depends on the immediately preceding state and only on it. A Markov process whose state space S is discrete is called a Markov chain. A Markov chain is called a discrete parameter Markov chain or a continuous parameter Markov chain according as the parametric space is discrete or continuous. Mathematically, A stochastic process {Xn ; n ≥ 0} is called a Markov chain, if for every xi S, Pr{Xn=xn/Xn-1=xn-1,………X0=x0}= Pr{Xn=xn/Xn-1=xn-1} (1.1) provided L.H.S. is defined. This equation indicates a kind of dependence between random variables Xn; intuitively, it implies that given the present state of the system, the future is independent of the past. The probability of moving from one state to another or remaining in the same state during a single time period is called the transition probability. The transition probability Pr{Xr+n= k/ Xr =j} from state j to state k in n steps (from state j in steps r to state k in step r+n) is called n step transition probability. We denote Pjk(n) = Pr {Xr+n = k/Xr = j} so that Pjk (1) = pjk Define pjk (0) = jk A single server queue system can also be considered to be in two states idle or busy according to whether server is idle or busy. The idle and busy states alternate and together constitute a cycle of an alternating renewal process. A busy period starts as soon as a customer arrives before an idle server and ends at the instant when server becomes free. 12 The case of the alternating renewal process can be generalized to cyclical movement of more than two states. A Markov renewal process becomes a Markov process when the transition times are independent exponential and are independent of the next state visited. It becomes a Markov chain when the transition times are all identically equal to 1. It reduces to a renewal process if there is only one state and then only transition times become relevant. Let {X(t)} be a Markov process with discrete state space & let the transition occurs at any instant of time ti , i=0, 1, 2……….. Then the sequence {Xn=X(tn+1)} forms a Markov chain and the time interval (tn+1-tn)=Δtn between transition have independent exponential distribution , the parameter which may depends upon Xn. Suppose that the transitions {Xn} of the process X(t) constitute a Markov chain but the transition intervals (tn+1-tn) may have any probability distributions with parameters which may depends not only on Xn but also on Xn+1. The process {Xn, tn} as well as {Xn, tn+1-tn} is called a Markov renewal process and the process {X(t)} is known as semi Markov process. More clearly semi Markov process is a stochastic process in which changes of state occurs according to Markov chain and in which the time interval between two successive transitions is a random variable whose distribution may depends upon the state from which the transition take place as well as on the state to which the next transition take place. e.g. The problem of occupational mobility, where transition between occupation as well as intervals between transitions is taken into account may be treated as a semi Markov process. Semi-Markov processes are used in the study of certain queuing systems. (b) POISSON PROCESS The Poisson process is a stochastic process with infinite number of states. It is applied to a system when the changes are independent of time. An infinite sequence of independent events occurring at an instant of time form a Poisson process, if the following conditions satisfied: (i)The total number of events in anytime interval X does not depend on the event which has occurred before the beginning of the period i.e. the number of arrivals in 13 non-overlapping interval are statistically independent and the process has independent and identically distributed increments. (ii) The probability of an event occurring in a small time interval t is t+ O ( t)2 where is some constant and O ( t) 2 means some function of t of order ≥ 2, i.e. ( ) ( ) t f t 0 as t 0. Here f ( t) is a function of t. (iii) Two or more units cannot arrive or serviced at the same time i.e. the probability of the occurrence of more than one event in the interval t and t + t is of the order O ( t) which is negligible. The Poisson process is a continuous parameter Markov chain and is widely used in queuing theory. It is well known that these conditions lead to the Poisson distribution. An important relationship exists between the Poisson and exponential distributions. This is clear from the statement of following theorem: Theorem: (Arrival Distribution Theorem) If the arrivals are completely random, then the probability distribution of number of arrivals in a fixed interval follows a Poisson distribution i.e. Pn(t) = ! ( ) n t e n t . Where t (the inter- arrival time) obeys the negative Exponential Law whose probability density function is defined as, f(t) = e - t , t >0. 1.7.1.1 RELEVANT PROBABILITY DISTRIBUTION FUNCTIONS Here, we review some probability distributions considered to be relevant to queuing theory. These are the Poisson distribution, the exponential distribution, the Erlang distributions and the geometric distribution, which play an important role in queuing theory. These distributions are described as follows: 14 1.7.1.1(a) POISSON DISTRIBUTION: The discrete probability distribution obeying the probability lawP(X =x) = x! e x x = 0,1, 2, … …. is called a Poisson distribution with parameter . The mean & variance of a Poisson distribution is given by . 1.7.1.1(b) EXPONENTIAL DISTRIBUTION: If X and X + X are the time of two successive arrivals then the probability that an arrival will take place in the interval (X, X + X) is given byf (x) P[X x X X] otherwise e x x 0, , 0 Where >0 is known as the prescribed parameter of the distribution. The mean & variance of an exponential distribution is given by 2 1 1 and respectively. Exponential distribution possesses a memory less or Markovian property and is the only continuous distribution to possess this property. It may be stated as follows. “If the interval between two occurrences is exponentially distributed, then memory less property implies that the interval to the next occurrence is statistically independent of time from last occurrence and has exponential distribution with same mean say 1/λ”. 1.7.1.1(c) REGULAR DISTRIBUTION: The time interval between two successive arrivals is assumed to be constant and the distribution is given byfor x a for x a f x 1, 0, ( ) ,where a being some constant. 15 1.7.1.1(d) ERLANG DISTRIBUTION: If f (x) is the probability density function that an arrival takes place in the time interval (X, X X) then for this distribution- ( 1)! ( ) ( ) 1 k x e f x k k x Where k & are some positive constant with k ≥ 1. The Erlang distribution is derived by combining a number of independent exponential variants. For k = 1, the distribution becomes an exponential distribution and when k = , it becomes a regular distribution. The inter-arrival time follows a Markov process which states that at any instant the time until the next arrival occurs is independent of the time that has elapsed since the occurrence of the last arrival. 1.7.1.1(e) GEOMETRIC DISTRIBUTION: A discrete random variable X is said to process a Geometric Distribution over the range x = 1, 2, … … … , if it assumes only the non-negative values and obey the probability distribution lawP(X = x) = 1 (1 ) x p p ; x = 1,2, … … … The mean and variance for the geometric distribution are p 1 and 2 1 p p respectively. The geometric distribution is sometimes known as “waiting time distribution” in a queuing system. The following table gives the collective information about different probability distributions- 16 SrNo Name of Distribution Range Parameters PDF Means Variance 1. Binomial x=0,1…n n=1,2… 0 p 1 ( ) (1 ) n x n x x p x C p p np np(1 p) 2. Poisson x=0,1,2… > 0 ! ( ) x e P x x 3. Normal - x - σ 2 > 0 2 ( ) 2 2 1 x e f x 2 4. Geometric x 1,2... 0 p 1 1 ( ) (1 )x P x p p p 1 2 (1 ) p p 5. (-)ve Binomial x=r,r+1... r = 1,2, … 0 p 1 x r x r Cr p p P x (1 ) ( ) 1 1 2 p r 2 (1 ) p r p 6. Erlang x ≥ 0 > 0, k≥1 ( 1)! ( ) ( ) 1 k x e f x k k x k 2 k 7. (-)ve Exponential x ≥ 0 > 0 x f (x) e 1 2 1 8. Gamma x ≥ 0 r x e dx r x e f x r x r r x 0 1 1 ( ) ( ) ( ) ( ) r 2 r 9. Beta 0 x 1 r > 0, s > 0 ( , ) (1 ) ( ) 1 1 r s x x p x r s r s r ( ) ( 1) 2 r s r s rs 17 1.7.2 FUZZY UNCERTAINTY: Let us suppose there is a research project or paper which is to be accepted by the authority or the editor. It is sent to the experts for the independent advice or comments about the acceptance. Each expert is asked to grade the project or the research paper to grade the product at some point in the scale of 0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0, where grade 0 means that the project is completely unacceptable and there is no uncertainty in the mind of the expert about it and grade 1 means that the project is completely acceptable and again there is no uncertainty in the mind of the expert about it. Grade 0.1 means that the project is almost unacceptable but there are some good features in the project which creates some uncertainty in the mind of the expert. Similarly grade 0.9 means the project is almost acceptable but there are some undesirable features which creates some uncertainty in the mind of the expert about its acceptability. Now when the authority or the editor receives the experts opinions, his concern about grade 0 and 1 gave him a clear indication but grade 0.1 and 0.9 creates a certain degree of uncertainty in his mind which is the same in both cases. Similarly grades 0.2 and 0.8 and 0.3 and 0.7, 0.4 and 0.6 represents the same degree of uncertainty form the authority or the editor. We call this type of uncertainty a fuzzy uncertainty as distinguished from probabilistic uncertainty. It is true that the fuzzy uncertainty is maximum when the expert gives the grade 0.5 as the authority is complete uncertain whether the project is accepted or rejected. 1.7.2(a) INTRODUCTION TO FUZZY SET THEORY A fuzzy set à is defined by à = {(x, à (x)): x X, Ã(x) [0, 1]}. In the pair {(x, Ã(x))}, the first element x belongs to the classical set A, the second element Ã(x), 18 belongs to the interval [0, 1], called membership function or grade of membership. The membership function is also a degree of compatibility or a degree of truth of x in Ã. Fuzzy set has been introduced by Zadeh L.A. [1965] as an extension of the classical notation of set. Classical set theory allows the membership of the elements in the set in binary terms; a bivalent condition an element either belongs or does not belong to the set. Fuzzy set theory permits the gradual assessment of the membership of elements in a set, described with the aim of a membership function valued in the real unit interval [0,1]. Fuzzy set theory is an extension of classical set theory where elements have degrees of membership. A fuzzy number is a quantity whose value is imprecise, rather than exact as is the case with “ordinary” (single-valued) numbers. Any fuzzy number can be thought of as a function whose domain is a specified set (usually the set of real numbers, and whose range is the span of nonnegative real numbers between, and including, 0 and 1000. Each numerical value in the domain is assigned a specific “grade of membership” where 0 represents the smallest possible grade, and 1000 is the largest possible grade. In many respects, fuzzy numbers depict the physical world more realistically than single- valued numbers. Suppose, for example, that we are driving along a highway where the speed limit is 55 km an hour (kmph). We try to hold our speed at exactly 55 kmph, but our car lacks “cruise control,” so our speed varies from moment to moment. If we graph the instantaneous speed over a period of several minutes and plot the result in rectangular coordinates, we will get a function that looks like one of the curves shown below: 19 fig 1.3 triangular fuzzy numbers fig 1.4 trapezoidal fuzzy numbers fig1.5 bell- shaped fuzzy number These three functions, known as membership functions, are all convex (the grade starts at zero, rises to a maximum, and then declines to zero again as the domain increases). However, some fuzzy numbers have concave, irregular, or even chaotic membership functions. Fuzzy numbers are used in statistics, computer programming, engineering (especially communications), and experimental science. The concept takes into account the fact that all phenomena in the physical universe have a degree of inherent uncertainty. 0 200 400 600 800 1000 1200 40 45 50 55 60 65 0 200 400 600 800 1000 1200 40 45 50 55 60 20 1.7.2(b). TRIANGULAR FUZZY NUMBER The fuzzy set à = (a1, a2, a3), where a1 < a2 < a3 and defined on R, is called the triangular fuzzy number, if the membership function of à is given by Ã(x) 0, otherwise. (a )/(a a ), a a , ( a )/(a a ), a a , 3 3 2 2 3 1 2 1 1 2 x x x x Defuzzification of TFN= 4 a 2a a 1 2 3 1.7.2(c) Trapezoidal Fuzzy Number A trapezoidal fuzzy number à = (a1, a2, a3, a4) is defined by the membership function as Ã(x) = 0 otherwise a a (a a ) (a ) 1 a a a a (a a ) ( a ) 3 4 4 3 4 2 3 1 2 2 1 1 x x x x x Defuzzification of Trapezoidal Fuzzy Number = 6 a 2a 2a a 1 2 3 4 1.7.2(d). α – LEVEL FUZZY INTERVAL For 0 ≤ α ≤ 1, the fuzzy set [aα, bα] defined on R is called α-level fuzzy interval if the membership function of [aα, bα] is given by α(x)= 0, . , [a , b ] otherwise x 21 1.7.2(e). α-CUT Let à be a fuzzy set on R and 0 ≤ α ≤ 1. Then α-cut of a Fuzzy set à is the crisp set A that contains all the elements of universe of discourse X whose membership grades in A are greater than or equal to the specified value α. i.e., Aα = {x X|µA (x) ≥ α}, α [0, 1] i.e. The α-cut Aα of fuzzy set à consists of points x such that µA(x) ≥ α, 1.7.2(f) THE FUZZY ARITHMETIC OPERATIONS UNDER FUNCTION PRINCIPLE Function principle proposed by Yager‟s [1997] and Chen, S.H., and Hsieh, C.H. [1999], defines fuzzy arithmetical operations for triangular and trapezoidal fuzzy numbers (Chen S.H. et al [2005] used these operations in their paper). We define some fuzzy arithmetical operations under Function Principle as follows: Let à = (a1, a2, a3) and ~ B = (b1, b2, b3) be two TFN, then the arithmetic operation on à and ~ B are given as follows: Addition à + ~ B = (a1+b1, a2+b2, a3+b3) Subtraction à - ~ B = (a1-b3, a2-b2, a3-b1) Multiplication Ã* ~ B = (a1b2+a2b1-a2b2, a2b2, a2b3+a3b2-a2b2) Division à / ~ B = (2a1/b1+b3, a2/b2, 2a3/b1+b3) provided à and ~ B are all non-zero positive TFN. 22 Next suppose à = (a1, a2, a3, a4) and B ~ = (b1, b2, b3, b4) are two trapezoidal fuzzy numbers. Then fuzzy arithmetic operations on them are given by: (1) The addition of à and B ~ is à + B ~ = (a1 + b1, a2 + b2, a3 + b3, a4 + b4) Where a1, a2, a3, a4, b1, b2, b3 and b4 are any real numbers. (2) The multiplication of à and B ~ is à B ~ = (C1, C2, C3, C4) Where T = {a1b1, a1 b4, a4b3, a4b4} and T1 = {a2b2, a2b3, a3b2, a3b3} C1 = min T, C2 = min T1, C3 = max T1, C4 = max T If a1, a2, a3, a4, b1, b2, b3, and b4 are all zero positive real numbers, then à B ~ = (a1b1, a2b2, a3b3, a4b4) (3) B ~ = (-b4, -b3,-b2, -b1) then the subtraction of à and B ~ is à B ~ = {a1 – b4, a2 – b3, a3 - b2, a4 – b1} where a1, a2, a3, a4, b1, b2, b3 and b4 are any real numbers. (4) 4 3 2 1 1 b 1 , b 1 , b 1 , b 1 B ~ B ~ 1 Where b1, b2, b3, b4 are all positive real numbers. If a1, a2, a3, a4, b1, b2, b3, and b4 are all nonzero positive real numbers then the division of à and B ~ is à / B ~ = 1 4 2 3 3 2 4 1 b a , b a , b a , b a (5) Let α R, then (i) α ≥ 0, α à = (α a1, α a2, α a3, α a4) (ii) α 0, Ã= (α a4, α a3, α a2, α a1) 23 1.7.2(g). GRADED MEAN INTEGRATION REPRESENTATION METHOD Chen, S.H., and Hsieh, C.H. [1999], introduced Graded mean Integration Representation Method based on the integral value of graded mean h-level of generalized fuzzy number for defuzzifying generalized fuzzy number. Here, we first define generalized fuzzy number as follows: Suppose à is a generalized fuzzy number. It is described as any fuzzy subset of the real line R, whose membership function µÃ satisfied are following conditions. 1. µÃ (x) is a continuous mapping from R to [0, 1], 2. µÃ (x) = 0, -∞ < x ≥ a1, 3. µÃ (x) = L(x) is strictly increasing on [a1, a2], 4. µÃ (x) = WA, a2 ≤ x ≤ a3, 5. µÃ (x) = R(x) is strictly decreasing on [a3, a4], 6. µÃ (x) = 0, a4 ≤ x < ∞ where 0 < WA ≤ 1 and a1, a2, a3 and a4 are real numbers. This type of generalized fuzzy numbers are denoted as à = (a1, a2, a3, a4; WA)LR. When WA =1, it can be formed as à = (a1, a2, a3, a4; WA)LR. Second, by Graded Mean Integration Representation Method, L-1 and R-1 are the inverse functions of L and R respectively and the graded mean h-level value of generalized fuzzy number à = (a1, a2, a3, a4 ; WA)LR is given by ( ) ( ) 2 1 1 L h R h h 24 Then the graded mean integration Representation of P(Ã) with grade WA, is (1.2) Where 0< h≤ WA and 0< WA≤ 1. Throughout this model, we use popular trapezoidal fuzzy number as the type of all fuzzy parameters in fuzzy production inventory models. Let B be a trapezoidal fuzzy number and be denoted as B= (b1, b2, b3, b4). Then we can get the Graded Mean Integration Representation of B by the formula (1.2) as P( B ~ ) = hdh b b h b b b b dh h 1 0 1 4 2 1 4 3 1 0 2 = 6 b1 2b2 2b3 b4 (1.3) 1.8 HISTORICAL DEVELOPMENT OF QUEUING THEORY/ LITERATURE SURVEY: The origin of queuing theory dates back to 1909, when Erlang A.K. published his fundamental paper on congestion in telephone traffic. In the development of queuing theory it is possible to distinguish four stages of different length and different character. The first stage is the time of intensive activity and represents the rise and fall of queuing theory. The second stage represents revival and separation. It splits queuing theory into two main types according to abstract or practical interest. The name of Erlang A.K. [1909], Crommelin C.D. [1932], comes immediately to mind in first stage. Felix Pollaczek, whose papers appeared in 1930, was the first to obtain the complete solution for the GI/G/s queue. hdh L h R h dh h A A w w 0 1 1 0 ( ) ( ) 2 P(Ã) 25 The application of stochastic processes marks the beginning of the second period in development of the queuing theory which may be called a period of growth, lasting approximately from the 1950s to the late 1975s. The year 1949 may be regarded as the turning point because at this time a birth and death process which was soon recognized as fundamental in queuing theory was introduced. The fact that queuing theory was recognized as a branch of the theory of stochastic processes resulted in its rapid development. Kendall [1951, 1953] was the pioneer who viewed and developed queuing theory from the perspective of stochastic processes. Jackson [1954] was the first to consider queues in series as a model of queuing system for the overhaul of aircraft engines, where stages of overhaul involve successive operation such as stripping, inspecting, repairing, assembling and testing. Every unit is served at each of stages called phases one after other. One of the most powerful relationships in queuing theory was developed by Johan D.C. Little in the early 1961s. Little related the steady-state mean system sizes to the steady-state average customer waiting times. Little‟s formulas areL = W and Lq = Wq Or Equivalently, Ws = Wq + 1 Ws → mean waiting time in system. Wq→ mean waiting time in queue. First, we briefly review the development of queuing networks models due to the fact that nearly all the multistage queuing networks models are either extensions of these or at least based on them. In various ways as queues in series, in bi-series, as 26 per their parameters priority queues, bulk queues, feedback queue, retrial queue, fuzzy queue etc. 1.8.1 QUEUES IN SERIES AND QUEUE NETWORK: The Queue Network is current area of great research and application interest with many extremely difficult problems. Networks of queues can be described as a group of nodes (say k of them) where each node represents a service facility of some kind with, let us say, Ci servers at node i, i = 1, 2, 3…….k. In the most general case, customers may arrive from outside the systems at any node and may depart from the system from any node. Thus customers may enter the system at some node, traverse from node to node in the system, and depart from some node, not all customers necessarily entering and leaving at the same nodes, or taking the same path once having entered the system. Customers may return to nodes previously visited, skip some nodes entirely, and even choose to remain in the system forever. 1.8.1 (a) CHARACTERISTICS OF QUEUING NETWORK 1. Arrivals from the outside to node i follow a Poisson process with mean rate λ1. 2. Service (holding) times at each channel at node i are independent and exponentially distributed with parameter (a node‟s service rate may allowed to depend on its queue length). 3. The probability that a customer who has completed service at node „i‟ will go next to node „j‟ (routing probability) is rij (independent of the state of the system), where i = 1, 2, 3…………….k, j = 0, 1, 2, 3……0..ki and ri0 indicates the probability that a customer will leave the system from node i. These properties are called Jackson Networks (Jackson [1957, 1963]). 27 1.8.2 TWO STAGE QUEUING NETWORKS: Tandem queuing system in which the output from one queuing process serves as the input to another, have been studied by many authors. The first work on sequences of queues in tandem starts with the work of Jackson [1954, 1956]. Jackson [1954] studied on open queue in tandem system with a finite number of customers and derived the differential-difference equations characterizing the system of two queues in series and obtained the steady-state solution: 1 2 0,0 0,0 1 2 ( ) , 1 1 1 2 1, 2 P P where P n n n n (1.4) The average number of customers in the system is 2 2 1 1 1 1 (1.5) The average number of customer waiting for service in the system (not including those in service is) 2 2 2 1 2 1 1 1 (1.6) Probability that there are n customers in Phase one is 1 1 1 n (1.7) Probability that there are n customers in phase two is 2 2 1 n (1.8) 28 Jackson was considered the first to introduce the concept of “product form” in the theory of queues. There are real life situations where the arriving customer passes successively through several distinct phase service channels in order to complete his service. O‟brein [1954], Hunt [1955] tackled the problem of queue in series taking Poisson input and exponential holding time. The queues in series is illustrated in the following figure Arrival Queue1 Queue 2 Departure Fig1.6 The output of a queuing system with Poisson input and exponential service time also Poisson, and we can generalize this result to any number of stages. It follows that the expected total waiting time in the system does not depend on the ordering, Riech [1957, 1963]. Starting with Jackson [1957] some of these results were extended to more complicated networks of queues. Suzuki [1963], studied a queuing system consisting of two servers in tandem with infinite number of customers before the two servers. The inter-arrival times are identically distributed, mutually independent random variables with distribution function x F(x) 1 e , if x ≥ 0 (1.9) 0 , if x < 0 The service times both stations are mutually independent. He derived the queue size and the waiting time distributions at the second server. He also considered a tandem queuing model with blocking where he studied a Markov chain imbedded 29 in the process. The transient solutions of the above models were obtained by Prabhu [1967]. Tandem queues with blocking can be studied in terms of an imbedded Semi-Markov process, Neuts [1968]. Weber [1983] considered the problem when there is a single server at each station and the service times at all stations are exponentially distributed. Based on a Laplace-transform method he showed that for a tandem system with infinite queues between the stations, the departure process is stochastically the same when the order of any two stations is interchanged. In this period of intensive activity, new methods were introduced, solution to problem of interest obtained, and many open problems suggested. The third period of queuing theory began in early 1975. This was a period of further extensive developments; it was also a period of maturity. In this period, queuing theory proceeded gradually from individually systems towards general consideration of types of stochastic processes and formulation of laws and methods applicable to sufficiently large classes of system. Rosenshine and Chandra [1975] developed approximate solutions for a two-stage tandem system and obtained the average steady-state queue length in four different but related tandem queues that arise in connection with the service system of an air terminal complex. Maggu et al. [1981] studied the “Transient analysis of serial queues under service parameter constraints”. The work was further extended by Singh T.P. [1994], Singh T.P. et al. [2007]. Recently Singh T.P. & Kumar Vinod [2007] made an attempt to have a “linkage of a scheduling system with serial and bi-serial queue network”. 1.8.3 MULTISTAGE QUEUING NETWORKS: In the following we will divide queuing networks theory into two major groups and we trace here the directions and development of the theory in the last few years. 30 1.8.3(a) OPEN NETWORKS SYSTEMS: In an open system, customers enter and are being serviced at various stations and leave the system. Reimaan, M.I. [1984] applied open queue network in heavy traffic. Research on using queuing networks in modeling a number of complex systems such as manufacturing, computer and communication networks and the evaluation of performance measures of networks and the design of such systems has grown very rapidly in the last few years. Lee and Zipkin [1992] considered a tandem multistage production process where demand is a Poisson process and the unit production times are exponentially distributed. Markovian queuing networks have been studied by many authors. Brandon & Yechiali [1991] studied a Markovian tandem queuing network with feedback (with Poisson arrivals, Exponential service times, and Bernoulli routing). Bambos & Wasserman [1994] studied a stationery tandem queuing networks with job feedback. Frenk et al. [1994] presented two algorithms for server allocation problems in manufacturing queuing network. For restricted networks, blocking may occur due to restrictions on each queue length, resulting in a decrease in maximum utilization. Queuing networks with blocking have proved useful in modeling stochastic systems such as computer systems, manufacturing systems, telecommunication system, and production lines, and quality control systems with a sequence of inspection stations. During the last three decades researcher concentrate on the approximation techniques for the solutions of queuing problems and in particular to queuing networks. Brandwajn and Jow [1988] proposed an approximation method for the solution of tandem queues with blocking. In his paper, Altiok [1989] studied a system of tandem queues with phase-type service times and finite buffers. He considered three cases: (1) the first queue has an infinite buffer; (ii) the first queue 31 has finite buffer; (iii) the first server is always busy. He used an approximation method to compute the steady-state probability distributions of the number of customers in each queue. This method decomposes the system into individual queues with revised arrival & service rates and with queue capacities. Then each queue is treated separately with an iterative scheme which relates these queues to each other. Lee & Pollock [1989] proposed an approximation method for analyzing the general configuration of a queuing network with blocking. In their paper, Song & Takahs [1991] studied problems arising in applications of the cross aggregation method to tandem queuing systems with production blocking. 1.8.3(b) CLOSED NETWORKS SYSTEMS: In a closed network a fixed and finite number of customers circulate through the network and no arrivals or departures are allowed. A survey of queuing networks theory literature reveals that some consideration has been devoted to the study of closed systems. Most of the published paper concentrated on the study of the two-stage model, where a customer upon completion of his service in one of the two stations can enter service immediately on the other, or he has to wait if there is a queue in front of him. Kumar Shanti & Yao [1989] studied the equilibrium behavior of the queue lengths in a product form closed queuing network when the service rates at a subset of stations are increased. Signman [1989] discussed the stability of closed queue network. Multiclass closed queuing networks was studied by Chevalier and Wein [1993] who considered the problem of finding an optimal dynamic sequencing polity to maximize the mean throughout rate. They generalized the results of Horrison and Wein [1990] from the system containing two-station network to a system containing a finite number of stations. 32 Closed queuing networks with blocking have been studied by many authors. Suri & Diehl [1986] studied a single class closed queuing networks with exponential service times and blocking. They used an approximation technique based partly on analysis and partly on heuristic arguments. In the paper, Van Dijik and Lamond [1988] investigated a new methodology to obtain simple and computationally attractive bounds for non-product form queuing systems to be applied to finite single-server exponential tandem queues. Their methodology is based on modifying the original system into product form systems that provide bounds for some performance measures. They studied a two stage tandem queue with blocking. Akyildiz [1988] presented an algorithm for throughput analysis of closed restricted systems. He states that if two queuing networks, one with blocking and the other without blocking, have an almost equal number of states, then their throughputs are almost equal. For an excellent review paper for closed systems with blocking, see also Onvural [1990]. 1.8.4 BI-TANDEM /BI-SERIES QUEUING SYSTEM Queues in bi-series were first studied by Maggu [1970]. While studying a queuing system having two queues Q1 and Q2 with servers S1 and S2, there are situation where the output from Q1 or the part of it constitute the input to Q2 and vice versa i.e. the output from Q2 or part of it forms the input to Q1 such an arrangement of two queues in series is termed as queues in bi-series/bi-tandem. This concept was introduced by Maggu [1970]. In brief we discuss the original model in bi-series as follows. Poisson arrivals from outside the system occur with mean rates 1 and 2 and join the respective queues Q1 and Q2 before the servers S1 and S2. It is assumed that the service time distributions for the queues Q1 and Q2 are mutually independent, 33 S1 negative exponentially distributed with respective parameters 1 and 2. Further it is assumed that p1, p2 (or p1 + p2 =1) denote respectively the probabilities. If a unit completing services at S1 to either leave the system or enter the subsequent service channels S2. Also q, q2 ≥0 (or q1 + q2 = 1) be the respective probabilities a unit having received service at S2 to either depart from the system or go through the subsequent phase channels. Maggu [1970] obtained the following differential difference equation P‟m,n (t) = -( 1 + 2 + 1 + 2)Pm,n(t) + 1Pm-1,n (t)+ 2Pm,n-1 (t) + 1p1 Pm+1,n(t) + 1p2Pm+1,n-1(t) + 1q1Pm,n+1(t) + 2q2Pm-1,n+1(t) for m, n > 0 (1.10) By using generating function technique (g.f.t), the expression has been obtained for the expected queues length of the system in study state: μ (1 p q ) (λ λ q ) λ λ q μ (1 p q ) (λ λ q ) λ λ q E(Q) 2 2 2 2 1 2 2 1 2 1 1 1 1 2 1 1 2 2 (1.11) m 1 Q1 p1 P2 q2 2 n Q2 S2 q1 Fig. 1.7 The study of Bi-serial queuing networks was further extended by Singh M [1979] and Singh T.P. [1986, 2005, 2007, 2009] Further Singh T.P., Gupta Deepak [2007, 2008] developed a network of bi-serial queues by taking two sub-systems in bi- 34 series with multi input source under service parametric constraints. Singh T.P. & Kumar Vinod [2006, 2007] studied bi-serial queue network under different situations. Singh T.P. & Maggu P.L. [2007] made an attempt to have a linkage of queues in series and bi-series with scheduling system. The work was further developed by Singh T.P. and Kumar Vinod. 1.8.5 CYCLIC/FEED BACK QUEUE MODELS Ko‟einsberg [1958] worked out a coal cutting problem and studied a cyclic queuing system consist of several service channels arranged in series. In some multistage queuing processes recycling or feedback may occur. Recycling is common in manufacturing processes, where quality- control inspections are performed after certain stages and parts that do not meet quality standards are sent back for reprocessing. Similarly telecommunication network may process messages through a randomly selected sequence of nodes with the possibility that some messages will require rerouting on occasion through the same stage. A multistage queuing system with some feedback is depicted in fig. Arrivals Departure Fig. 1.8 If we consider a closed network of k nodes such that 0 ( ) 1, ( , 1) 1, ( 1,1 1) elsewhere i k j j i i k rij (1.12) Service Service Service 35 Then we have a cyclic queue. A cyclic queue thus is a sort of series queue in a „circle‟ where the output of the last node feedback to the first node. A cyclic queue or feed-back model can be supposed a special case of a closed queue network; we apply General Jackson Network result for single server at each node and get a present form solution k k n k n n n n n n P cρ ρ ............ρ 1 2 1 , 2 3 ,......... 1 2 (1.13) With k j i i j ji j r 1 Using (1.13) in traffic equation result is iρi = ( 1) μ ρ (i 1,2,........k) i 1 i 1 i x x (1.14) (1.15) i.e. k k k 1 1 3 1 3 3 2 4 3 2 1 3 2 1 2 , , ,................. The machine repair problems, can be supposed two node cyclic queues, with node 1 representing machine “up” node and the numbers of server at this node being the number of machines desired operational, denoted by on the presence of a queue at this node means that spare machines are on hand, while an idle server represents fewer that M machine operating. The mean “service” (holding) time at this node per customer (machine) is 1 where machine mean failure rate is. In their paper, Horne and Irony [1994] studied the ship-to-shore transfer of cargo ship that is (i 1,2,3........k) i 1 ρ μi i 1 μ (i 1) k ρ 1 μ k μ i ρ 36 located offshore. In such situations cargo is transferred from ships to smaller craft, which in turn transport the cargo the remaining distance to shore. These crafts cycle back and forth from the ship to the shore until the transfer is complete Singh T.P. [1994, 1996] studied the steady state behavior of a feedback Queue Model in a wider sense. The work in feedback was further extending by Kusum, Gupta Deepak & Singh T.P. [2010]. In this thesis we have developed the generalized feedback queue modes. 1.9 BULK QUEUES: Letters arriving at a post office, ships arriving at a port in convoy, people going to a theatre or restaurant, and so on, are some of the examples of queuing situations in which customers do not arrive singly, but in bulk or groups. Also, the size of an arriving group may be a random variable or a fixed number. The study of bulk arrival queues may be said to have began with Erlang‟s solution of the M/Ek/1 queue (Brockmeyer etal. [1948]). The major contribution on bulk arrival queues are made by many researchers. Bulk service queue models have been analysed by many authors. Neutts [1967], Borthakur [1971], Baily [1954], Aftabh Begum [1996], Chaudhary G. [2002], Laxami & Srinivasan et al [2003], Medhi.j [1984a, 1984b], Indra and Sharda [2004], Ahmed [2007] considered a multi-channel bi-level heterogeneous servers bulk arrivals queuing system with Erlangian service time. Indira and Vijay [2010] analyzed a two dimensional bulk arrival queuing model with exhaustive and non-exhaustive service policy. A batch sizes queuing system provide practical models for performance evaluation in computer and communication system, e.g. multiprogramming computer systems, for which each program requires the loading of memory units from a main memory store; a circuit-switched telecommunication system which supports a 37 variety of traffic type such as voice, video etc. Each of which having different bandwidth requirements and holding-time distributions. 1.10 RETRIAL QUEUES: In retrial system, an arriving customer who finds the server busy is obliged to leave the service area and join a pool of unsatisfied customers called the „Orbit‟. From orbit, each customer reapplies for service after a random amount of time. The performance of retrial queuing system has been studied by several authors in terms of their main characteristics such as the limiting distributions, the busy period and the number of customers served during busy period (Choo and Conolly [1979], Falin & Artalejo [1998], Artalejo and Lopez [2001], Singh T.P. & Kusum [2012]). 1.11 RECENT/ FOURTH PERIOD OF DEVELOPMENT OF QUEUING THEORY: A survey of queuing networks theory literature reveals that relatively little consideration has been devoted to the application of the models. This is in fact due to the complexity of the models. Certainly there are some successes of the theory in areas such as manufacturing systems, production processes & computer systems. Also the application of queuing networks theory of different areas of current interest can be added to its successes. But in the Socio-political areas, the applications of queuing networks theory are still at the development stages. In this period the emphasis on application on queue models and queue model under fuzzy environment has been developed. Queuing network models have also been used to investigate the performance of store-and-forward packet-switched computer-communication networks, service centers are used to represent communication lines and concentrators as well as processors. Some of the problems that have been investigated involve topology 38 design, routing, capacity selection flow-control, congestion control and distribution of end-to-end delay (Reiser [1982], Srinivasa Rao et al [2000, 2006]). Communication networks may involve hundreds of service centers and many classes of customers (one for each logical communication path in the network). Exact solution of models of such networks is still feasible, however, by exploiting the fact that each class of customers typically visits only a small fraction of the service centers in the network. Mathematical analysis of queue models provides the basic frame work in communication network since a close resemblance can be seen between queue model and communication system. The message can be supposed as customers or jobs, communication buffers can be treated as the waiting lines and all the concerned activities in transmission of the message as services, capacity of buffers can be regarded as capacity of queue system while service discipline may be taken as FIFO depending upon situation. Application to computer systems and computercommunication networks has motivated extensive progress in the theory of queuing networks. Recently, Rao Srinivasa et al [2006] & Singh T.P. [2011] did the commendable work in this direction and applied the interdependent queue model with bulk arrivals to a communication network in order to reduce the average delay of transmission and variability of buffer content. Singh T.P. & Kusum [2012] studied a fuzzy queue model to interdependent communication system. In this thesis we have also examined the effect of covariance in the context of interdependence communication system under fuzzy situation. Siddhartan et al. [1996] analyzed the effect on patient waiting times when primary care patients use the Emergency Department. They proposed a priority discipline for different categories of patients and then a first in first out discipline for each 39 category. They find that the priority discipline reduces the average wait time for all patients: however, while the wait time for higher priority patients reduces, lower priority patients endure a longer average waiting time. Fiems et al. [2007] investigated the effect of emergency requests on the waiting times of scheduled patients with deterministic processing times. Fomundam S.F. & Herrmann J.W. [2007] provided sufficient information to analysts who are interested in using queuing theory to make a healthcare process and want to locate the details of relevant models. Heidmann [1997] used queuing theory to describe traffic behavior at signalized and unsignalized intersection. Vandaele, Woensel and Verbruggen [2000] used queuing theory to describe uninterrupted traffic flows. 1.11(a) FUZZY QUEUE MODELS: The concept of fuzzy set given by Zadeh L.A. [1965] applied in queuing theory and fuzzy queue models have been formed which have wider applications in real practical situation. The credit goes to “Buckley J.J. [1990], Lie and Lee [1989], Lie and Negy [1992], Kao [1999], Gunny and Ritha [2006], Pourdavish. A [2009], Singh T.P. [2009] etc. in fuzzy queue modeling, Fuzzy logic, Fuzzy arithmetic offers a better way of representing reality. In fuzzy logic, a statement is true to various degrees, ranging from completely true to half-truth to completely false. Ritha W. & Robert L. [2010] described a priority disciplined queuing models by using fuzzy set theory. They optimized a fuzzy priority queue models (preemptive priority, non-preemptive priority in which arrival rate, service rate are taken fuzzy numbers). Recently Fuzzy queue models have been further studied by Singh T.P. [2011, 12]. These authors analyzed fuzzy queues using Zadeh‟s extension principle. 40 Fuzzy queuing & fuzzy control of queues would be a new approach for surveying the queue system. If we let service time is expressed by possibility rather than probability the fuzzy queue method would be a more realistic approach than classical queuing theory methods. In our thesis, we have discussed the advance network of queue model with specific emphasis on feedback in addition a study of impatient customers has been made and the various Fuzzy Queue Model have been explored. 1.12 METHODOLOGY USED There are two basic approach to solve a queuing problem; mathematical approach and simulation approach. The mathematical approach makes use of probability distribution concept to represent arrival and service rates. The simulation approach is an iterative method wherein an approximate solution to the problem is obtained through simulated experiment based on random samples drawn from inter-arrival and service time distribution. Due to complexity of the models being considered exact solution techniques are not easy, most of the research works attempt at developing approximate solution techniques. These techniques differ in accuracy and vary in complexity, but most of them are based on the idea of decomposing a queuing network into individual queues with revised parameters; and after that the available or newly developed techniques are used to analyze the individual subsystems in isolation. Simulation can also used in solving such models. In our work, we have used mathematical approach forming differential difference equations of the system. The following, mathematical techniques have been widely used while solving the various queue models and to predict the behavior of the system. In our thesis we have applied most of these techniques. 1. Generating Function Technique 41 2. Matrix Solution 3. Laplace Transformation Technique & Rouche‟s theorem 4. Lagrange Auxiliary Equations 5. Probability Distribution Functions etc. 6. Fuzzy Triangular & Trapezoidal membership functions. 7. α-cut approach. 1.13 WORK DONE IN THE THESIS: Following the fundamental and historical development of the queuing theory in this chapter, the rest of the thesis is organized as: Chapter 2 deals with the transient as well as steady state behavior of a serial queue system with impatient customers and a case study has been explored on the basis of result phenomenon in a famous restaurant. This chapter is divided into three subchapters. Subchapter 2A is an attempt to study a system consists of two queues in series in which arrival follow Poisson distribution and the service rate and reneging rate depends upon their respective queue length. The expression for the system characteristics have been derived and the effect of customers impatience on the development of waiting lines has been analyzed for the model behavior whenever, queue discipline is first come first out. Subchapter 2B deals with the cost analysis of M/M/1/N queuing system with impatient customers. We assume that the arriving customers balk with some probability and renege follow negative exponential distribution. The Markov process techniques have been applied to develop the steady state probability. Matrix form solution is given along with some performance measures. The cost analysis of the model has been developed to find the optimal service rate. To demonstrate the use of the model, different parametric values have been specified and lucid illustration has been 42 presented with the help of table and graphs. Subchapter 2C deals with a case study regarding the application of queue model in a restaurant. This study aims to show that queuing theory satisfies the stochastic model when applied with a practical and real time scenario. We obtained the data from a famous restaurant. The various performance measures as utilization rate, waiting time in queue system, queue length and the probability of potential customers to balk based on the data have been studied using Little‟s formulae. The study in one way is helpful to increase the quality of service in the Restaurant by anticipating if there are many customers in the queue; on the other hand, the restaurant can set a target profit that should be achieved on daily basis. Chapter 3 provides a generalized stochastic analysis of a feedback serial queue model. The chapter studied a network of three queues in series with feedback. This chapter studies the steady as well as transient state behavior of a feedback queue network in which after service at a serial channel the customer/items has two choices either to proceed to next or to feedback to one preceding channel. The arrivals and departure both follow the Poisson law at each channel. The Generating function technique (g. f. t), L-Hospital rule, laws of calculus & statistical formulae of Marginal probability distribution has been applied as a solution methodology in order to find various queue parameters. The objective of the study is to find out the mean queue size of each channel and for the entire system. This chapter divided into two sections. Section 3.1 discusses the transient behavior of three stage feedback queue system as a specific case while section 3.2 is a general case with applications. Chapter 4 deals with a stochastic analysis of a complex system of feedback network queue model comprised of two bi-serial systems. The chapter discusses a complex Network of queues in which a common channel (called central channel) is 43 linked in a series with two sub-systems, one containing bi-serial channels while second sub-system consist of parallel channels. There is a feedback service from the centrally linked channel to each sub systems. All the activities in the system are performed under stochastic environment. The arrivals follow Poisson distribution at each channel and service times are distributed exponentially at each channel. The system performance characteristics have been found using statistical formulae, laws of calculus and generating function technique. The model finds its applications in various decision making process such as administrative office etc. In Chapter5 we discuss the various system performance measure of a machine interference queue model in fuzzy environment. Machine interference queue models play vital role in many manufacturing concern, textile industries, grinding machines, at service centers and Clint server computing problems. The system performance measures of these problems are breakdown rate, loss of production rate, service rate, etc. In real world situations these parameters may not be presented precisely due to uncontrollable factors i.e. the models behave fuzzy in nature. In this paper, a methodology for constructing system performance has been proposed in which break down rate and service rate are supposed trapezoidal fuzzy number. α- cut approach and fuzzy arithmetic operator are used to construct system characteristics membership function. The model finds suitable for practitioners. The validity of the proposed model has been presented through numerical illustration. Chapter 6 explores the application of an interdependence queue models to a communication system with the voice packetized, Statistical multiplexing under stochastic as well as in fuzzy environment. The chapter is divided into three sections. The first section 6.1 discusses the interdependence communication system in which the massage services and the transmission completion has been 44 assumed trapezoidal fuzzy in nature. Arrival process is bulk arrival and voice transmission massage is divided into voice packets of random lengths. Section 6.2 explores the application of an interdependence queue model to a communication system with the voice packetized statistical multiplexing. The performance of the statistical multiplexing is measured by approximating the arrival of packets and transmission under a bi-variate Poisson process, whit the assumption that data packets are stored for transmission and transmitted as when the transmitter is idle after voice transmission. The idle time of transmitter follows the negative exponential distribution. The steady state solution and the system characteristics are derived and analyzed for the model. Table and Graph are given for better understanding. The strategies for integrated data/voice transmission are discussed for channel capacity utilization more optimally. Section 6.3 is an extended work of 6.1 and 6.2. The applications of interdependence fuzzy queue model to a communication system have been discussed. The steady state solution and system characteristics has been derived & analyzed for the model using fuzzy arithmetic.