Chapter 17 WAITING LINES Queuing Theory. Mathematical approach to the analysis of waiting lines.

Examples: the lines at supermarket check-outs. Goal of Queuing Analysis To minimize total cost: those associated with customers waiting for service and those associated with capacity. Capacity costs are the costs of maintaining the ability to provide service. The costs of customer waiting include the salaries paid to employees while they wait for service, the cost of the space for waiting, and any loss of business due to customers refusing to wait and possibly going elsewhere in the future. To balance the cost of providing a level of service capacity with the cost of customers waiting for service. System Characteristics 1. Population source There are two possibilities: infinite-source and finite-source populations. Infinite source are customer arrivals are unrestricted. Finite source are the number of potential customers is limited. 2. Number of servers (channels) Channel is a server in a service system. Systems can be either single or multiplechannel. 3. Arrival and service pattern The commonly used models assume that the customer arrival rate can be described by a Poisson distribution and that the service time can be described by a negative exponential distribution. Poisson distribution often provides a reasonably good description of customer arrivals per unit of time. The negative exponential distribution often provides a reasonably good description of customer service times. 4. Queue discipline (order of service) Queue discipline is the order in which customers are processed. All but one of the models to be described assumes that service is provided on a first-come, first served basis. Measures of System Performance 1. Average number of customers waiting, either in line or in the system. 2. Average time customers wait, either in line or in the system. 3. System utilization, which refers to the percentage of capacity utilized. 4. Implied cost of a given level of capacity and its related waiting line. 5. Probability that an arrival will have to wait for service. Queuing Models: Infinite-Source All assume a Poisson arrival rate. The models pertain to a system operating under steady state conditions. Assume the average arrival and service rates are stable. The four models described are: 1. Single channel, exponential service time 2. Single channel, constant service time 3. Multiple channel, exponential service time 4. Multiple priority service, exponential service time Symbols Customer arrival rate Service rate Lq The average number of customers waiting for service Ls The average number of customers in the system (waiting and/or being served)

p Wq Ws I/ Po Pn M Imax

The system utilization The average time customers wait in line The average time customers spend in the system (waiting in line and service time) Service time The probability of zero units in the system The probability of n units in the system The number of servers (channels) The maximum expected number waiting in line

Basic Relationships 1. The average number being served is the ratio of arrival rate to service rate: r=/ 2. The average number in the system is the average number in line plus the average number being served: Ls = Lq + r 3. The average time in line is the system is the average number in line divided by the arrival rate: Wq = Lq / 4. The average time in the system is the sum of the time in line plus service time: Ws = Ws + 1 / 5. System utilization is the ratio of arrival rate to service capacity: p = / M Model 1: Single Channel, Exponential Service Time 1. Average number in line: Lq = 2 / ( - ) 2. Probability of zero units in the system: P0 = 1 ( / ) 3. Probability of n units in the system: Pn = P0( / )n Model 2: Single Channel, Constant Service Time 1. Average number in line: Lq = 2 / 2( - ) Model 3: Multiple Channel 1. Average number in line:

Lq = [ (/)M I (M 1)/M - )2]P0 2. Probability of zero units in the system: P0 = [ M-1n=0 (/)n/nI + ( / )M/MI(1 /M)]-1 3. Average waiting time for an arrival not immediately served: Ws = 1 - M - 4. Probability that an arrival will have to wait for service: Pw = [( / )M] P0/ M(1 /M)

Cost Analysis The optimal capacity is one that minimizes the sum of customer waiting costs and capacity or server costs. Minimize Total Cost = Customer waiting cost + Capacity Cost Capacity is incremented one unit at a time and the total cost is computed at each incremented. The computation of customer waiting costs is based on the average number of customers in the system.

Maximum Line Length Another question that often comes up in capacity planning is the amount of space allocate for waiting lines. n = log K / log p or ln K / ln p K = 1 Specified percentage / Lq(1 p) Model 4: Multiple Priorities Multiple-priority model. Customers are processed according to same measure of importance. 1. System utilization: P = / M 2. Intermediate values: A = /(1 p) Lq Bk = 1 - kc=1 /M (B0 = 1) 3. Average waiting time in line for units in kth priority class Wk = 1/ABk-1Bk 4. Average time in the system for units in the kth priority class W = Wk + (1/) 5. Average number waiting in line for units in the kth priority class Lk = k x Wk Revising Priorities If any of the waiting times are deemed too long by management, there are several options. One is to increase the number of servers. Another is to attempt to increase the service rate. Another approach is to reexamine the membership of each of the priority classification. 3 k=1kWk Queuing Models: Finite-Source In order to use the finite-queuing, follow this procedure: 1. List the values for: a. N, population size b. M, number of servers/channels c. T, average service time d. U, average time between calls for service per customer 2. Compute the service factor, X = T/(T + U) 3. Locate the section of the finite-queuing for N. 4. Using the value of X as the point of entry, find the values of D and F that correspond to M. 5. Use the values of N, M, X, D, and F as needed to determine the values of the desired measures of system performance. Symbols D Probability that a customer will have to wait in line F Efficiency factor: 1 percentage waiting in line H Average number of customers being served J Average number of customers not in line or in service L Average number of customers waiting in service M Number of service channels N Number of potential customers T Average service time U Average time between customer service requirements per customer W Average time customers wait in line X Service factor

1. Service factor: X = T/(T+U) 2. Average number waiting: 3. Average waiting time: 4. Average number running: J = NF(1 X) 5. Average number being served: 6. Number in population: N=J+L+H Other Approaches 1. The system may be currently in operation and indicated design changes may be too costly, or space restrictions may prevent changes from being made reasonably. 2. Particularly suited for queuing systems in which the customers are people rather than inanimate objects is to provide some form of diversion so that the waiting time becomes more tolerable. H = FNX L = N(1 F) W = L(T + U) / (N L) = T (1 F)/XF