Professional Documents
Culture Documents
Ks =
The usual physical interpretation of K is that the queueing system has only limited waiting room that can accommodate a maximum of K customers in the system (and K - s in the queue). For example, Herr Cutters barber shop described in Section 11.1 has only 3 chairs (other than the barbers chair) for waiting. Thus, since s = 1 in this case, K =4 and K - s = 3. Another possible interpretation of K is that arriving customers will leave whenever they find too many customers (K) ahead of them in the system because they are not willing to incur such a long wait. This behavior, referred to as balking, is quite common in commercial service systems. In the case of Herr Cutters barber shop, he has found that his customers typically will leave (balk) when they find that there already are 4 customers in front of them (including one currently receiving a haircut), so 4 is the appropriate value for K. Here is the model that is normally used for such queueing systems. Assumptions: Exactly the same as for the M/ M/s model (Section 11.6) except that when the system already is full with K customers, any new arriving customers leave without entering the system. Thus, the model is just the finite queue variation of the M/ M/s model.
CD S Ch 11-2 Lengthy formulas are available for obtaining the usual measures of performance (L, Lq, W, and Wq), as well as the Pn probabilities, for this model. Fortunately, an Excel template is available in your MS Courseware that includes a routine for calculating these measures of performance for you. For any given value less than 1 of the utilization factor, = /s, the values of L, Lq, W, and Wq for this model are less than the corresponding values for the M/ M/s model. The reason is that the finite queue turns some customers away so that the remaining customers encounter less congestion. Although the M/ M/s model requires < 1 to enable the queueing system to reach a steady-state condition, this is not necessary for the finite queue variation of the model. Any positive value of will still yield steady-state results because the servers are still able to keep up with the customers who actually enter the system. For example, consider Herr Cutters barber shop. As indicated in Section 11.1, the mean arrival rate (including customers who balk) for this queueing system is = 3 customers per hour, the mean service rate is = 3 customers per hour, and s = 1. Therefore, =
! 3 = = 1, s 1(3)
P0 = P1 = P2 = P3 = P4 = P5 = P6 = P7 = P8 = P9 = P 10 = P 11 = P 12 =
Figure 1
Probability
The Excel template for the finite queue variation of the M/ M/s model is applied here to Herr Cutters barber shop.
CD S Ch 11-3 Managers of commercial service systems (such as Herr Cutter) do not like to lose customers who arrive and then leave because the system already is full. Therefore, one probability of special interest with this model is PK = probability that the system is full = proportion of arriving customers that are lost. For example, since K = 4 for Herr Cutters barber shop, Figure 1 indicates that PK = 0.2 in this case. When designing a queueing system with a finite queue, one important decision often is how much waiting room to provide. The value of PK for various alternative values of K can help guide this decision. Since denotes the mean arrival rate for all customers (both those who enter the system and those who do not), P K = mean rate at which customers are lost (i.e., the expected number lost per unit time), = mean rate at which customers actually enter the system.
Consequently, this last quantity represents the true utilization of servers, rather than = /s, for this model. Because = 3, PK = 0.2, and = 1 for Herr Cutters barber shop, the mean rate at which he loses customers who dont want to wait for a haircut is PK = 0.6 customers per hour. The mean rate at which customers enter the barber shop and decide to wait for a haircut is (1 - PK) = 2.4 customers per hour. The average fraction of time that Herr Cutter is busy giving haircuts is (1 PK) = 0.8. Since (1 - PK) is the mean rate at which customers actually enter the system, this quantity needs to replace when applying Littles formula (L = W and Lq = Wq) for this model. For example, this formula becomes L = 2.4 W and Lq = 2.4 Wq (as holds in Figure 1) for Herr Cutters barber shop.
CD S Ch 11-4 population does affect the mean arrival rate in this way, because the number already in the system significantly affects how many are left who might arrive. Such a calling population is referred to as a finite calling population. For such a queueing system, let N = size of the calling population. In the case of the Dupit case study introduced in Section 11.4, each Tech Rep originally was assigned (approximately) 150 machines to be repaired as needed. Thus, the calling population for the queueing system corresponding to a one-person Tech Rep territory consists of these 150 machines, so N = 150. However, this is such a large number that the 150 machines are considered to be an infinite calling population, so the M/ M/1 model was used originally to analyze these territories. However, now consider a similar example where the calling population is so small that it must be considered finite.
EXAMPLE
1. A machine operator is responsible for running five semi-automatic machines. Each time a machine finishes a job, the operator must unload it and then set it up and load it for the next job (which is always available). This is the service being performed in a single-server queueing system, where the machine operator is the server and the five machines are the members of the calling population (N = 5). Even though the machines are stationary, each machine is considered to be a customer entering the queueing system when it first needs service, and it leaves the queueing system when service is completed. The time required to service a machine in this way has an exponential distribution with a mean of 1 hour (so the mean service rate is =1 service per hour). Upon completing service, the time until a machine needs service again has an exponential distribution with a mean of 5 hours (so the mean arrival rate of this machine to the queueing system when it has been running is 0.2 arrival per hour).
2.
3.
Because of this last assumption, the current mean arrival rate to the queueing system of machines needing service depends completely upon the number of machines that currently are already in the queueing system needing service (and so are not running). If all five already are in the queueing system (so none are running), then the current mean arrival rate must be 0. However, if none are in the queueing system (so all five are running), then the current mean arrival rate would be 5 (0.2) = 1 arrival per hour. Similarly, the current mean arrival rate is 0.8, 0.6, 0.4, and 0.2 arrival per hour when the number of machines in the queueing system (and so not running) is 1, 2, 3, and 4, respectively. This is an example of the finite calling population variation of the M/ M/s model. In this example, s = 1, but any number of servers is allowed. For the general model, assumption 3 of the example is rephrased as follows.
CD S Ch 11-5
3.
Upon completing service, the time until a member of the calling population needs service again has an exponential distribution with a mean of N/ (so the mean arrival rate of this member to the queueing system when it is not already there is /N). Thus, = current mean arrival rate to the queueing system when no members of the calling population already are there.
Except for this assumption replacing the usual assumption about interarrival times, the assumptions of this model are identical to those for the M/ M/s model. Using the Excel template for this model on the example gives the results shown in Figure 2.
A B C D E 1 Template for M/M/s Finite Calling Population Model 2 3 Data 4 ! = 1 (max arrival rate) 5 = 1 (mean service rate) 6 s= 1 (# servers) 7 N= 5 (size of population) 8 9 10 11 0.3 12 0.25 13 0 0.2 14 1 15 2 0.15 16 3 0.1 17 4 18 5 0.05 19 6 0 20 7 0 2 4 6 8 10 12 14 16 18 20 22 24 21 8 Number of Customers in System 22 9 23 10
Results L = 1.424339107 L q = 0.709206928 W= Wq = "= !-bar = P0 = P1 = P2 = P3 = P4 = P5 = P6 = P7 = P8 = P9 = P 10 = 1.991714468 0.991714468 1 0.715132179 0.284867821 0.284867821 0.227894257 0.136736554 0.054694622 0.010938924 0 0 0 0 0
Figure 2
Probability
The Excel template for the finite calling population variation of the M/ M/s model is applied here to the example where a machine operator is running semiautomatic machines.
Thus, the expected number of machines that currently need service is L = 1.424 (or Lq = 0.709 if a machine already in service is not counted), the expected waiting time of a machine to receive service is W = 1.99 hours (or Wq = 0.99 hour if the service time is excluded), the probability that 0 machines currently need service is P0 = 0.285, etc. One new type of result given by this template is
CD S Ch 11-6
! s
CD S Ch 11-7 Table 1 Comparisons of the Value of L for the M/M/s Model and Its Two Variations
Value of L M /M /s Finite Calling Population Variation N=5 1.03 1.27 1.41 2.14 2.37 2.49 N = 20 1.76 2.54 3.11 3.35 4.07 4.53
s 1 1 1 5 5 5
! s
Model Finite Queue Variation K=5 3 9 99 5.14 11.36 101.5 1.70 2.20 2.47 3.08 3.41 3.56 K = 20 2.95 6.42 9.63 5.05 8.30 10.92
Note that the only cases where the M/ M/s model provides a close approximation to either of its variations are for the finite queue variation with = 0.75 and K = 20. The approximation still is in the ballpark for the finite queue variation with = 0.9, K = 20, and for the finite calling population variation with = 0.75, N = 20. However, it is way off for the other cases. The lesson is that this approximation should not be used when is nearly 1 unless K or N is huge. Even with smaller values of , the M/ M/s model should not be used to approximate the finite queue variation unless K is at least several times as large as the value of L obtained from the M/ M/s model. Even larger values of N are needed before approximating the finite calling population variation by the M/ M/s model.
CD S Ch 11-8
= Mean
=0
for the exponential distribution for the degenerate distribution (constant service times)
For many queueing systems, the amount of variability in the service times falls somewhere between those for the exponential and degenerate distributions. Another service-time distribution that fills this middle ground is the Erlang distribution. (It is named after A. K. Erlang, a Danish mathematician in the early twentieth century whose work for the Copenhagen Telephone Company in analyzing the waiting of telephone calls began the development of queueing theory.) This distribution has a parameter k, called the shape parameter, that determines the standard deviation . In particular,
!=
1 mean k
where k is allowed to be any positive integer (k = 1,2,3, . . .). Figure 3 shows the shape of this distribution for several values of k. When k = 1, this figure indicates that the shape of the Erlang distribution is the same as for the exponential distribution. This is no coincidence, because the two distributions actually are the same for k = 1. For larger values of k, the figure shows that the high point of the distribution (called the mode) no longer is at 0. In fact,
CD S Ch 11-9
Figure 3: The shape of several Erlang distributions with the same mean but with different values of the shape parameter k.
Table 14.2 The Relationship Between the Standard Deviation and the Mean for Service-Time Distributions
Distribution Exponential Degenerate (constant) Erlang, any k (k = 1, 2, ... ) Erlang, k = 2 Erlang, k = 4 Erlang, k = 8 Erlang, k = 16 Standard Deviation mean 0
1 mean k 1 mean 2
1 mean 2
mean 2 2 1 mean 4
CD S Ch 11-10
!=
1 1 1 mean = k k
The ability to choose any positive integer value for k provides considerable flexibility to match in this model quite closely to the anticipated standard deviation for the queueing system under study. The procedures for obtaining the measures of performance for the M/Ek/s model are even more complicated than for the M/D/s model. However, considerable numerical results have been generated.1 Figure 4 compares the values of L for three values of k for the case of two servers (s = 2). The values of W, Wq, and Lq then can be obtained from L. (The percentage differences in the values of Lq = L - l/ between the cases of k = 1, k = 2, and k = 8 are considerably larger than these differences in L shown in the figure.)
Extensive tables and graphs for both the M/Ek/s and M/D/s models are available in F.S. Hillier and O.S. Yu, with D. Avis, L. Fossett, F. Lo, and M. Reiman, Queueing Tables and Graphs (New York 1981), Elsevier North-Holland.
CD S Ch 11-11
Figure 4: Values of L for the M/Ek/2 model for various values of the shape parameter k.
REVIEW QUESTIONS
1. 2. 3. 4. 5. 6. 7. 8. 9. What assumption is made by the finite queue variation that is different from those for the M/ M/s model? What is the significance of PK for the finite queue variation? What is the formula for the average fraction of time that servers are busy for the finite queue variation? What is meant by a calling population? A finite calling population? What assumption is made by the finite calling population variation that is different from those for the M/ M/s model? What is the formula for the average arrival rate in the long run for the finite calling population variation? What is the formula for the utilization of servers for the finite calling population variation? When is nearly 1, does the M/ M/s model usually provide a good approximation of either the finite queue variation or the finite calling population variation? Compare the M/M/s, M/D/s, and M/Ek/s models in terms of the amount of variability in the service times.
CD S Ch 11-12
Glossary
Calling population: The population of all potential customers who might need to come to a queueing system for service. Erlang Distribution: A common service-time distribution whose shape parameter k specifies the amount of variability in the service times. Figure 3 shows its shape. Finite calling population: A calling population that is small enough that the number of customers already in the queueing system significantly affects the mean arrival rate of new customers. Finite queue: A queue that can hold only a limited number of customers. Infinite queue: A queue that can hold an essentially unlimited number of customers. Queue: The waiting line in a queueing system. The queue does not include customers who are already being served. Queue capacity: The maximum number of customers that can be held in the queue.
Problems
An Excel template is available in your MS Courseware to assist you in doing most of the following problems. This is indicated by the letter E in front of each problem number or the relevant parts. E 1. Janet is planning to open a small car-wash operation, and she must decide how much space to provide for waiting cars. Janet estimates that customers will arrive randomly at a mean rate of 1 every 4 minutes, unless the waiting area is full, in which case the arriving customers will take their cars elsewhere. The time that can be attributed to washing one car has an exponential distribution with a mean of 3 minutes. Compare the average fraction of potential customers that will be lost because of inadequate waiting space if (a) 0 spaces (not including the car being washed), (b) 2 spaces, and (c) 4 spaces were provided. 2. George is planning to open a drive-through photo-developing booth with a single service window that will be open approximately 200 hours per month in a busy commercial area. Space for a drive-through lane is available for a rental of $200 per month per car length. George needs to decide how many car lengths of space to provide for his customers. Excluding this rental cost for the drive-through lane, George believes that he will average a profit of $2 per customer served (nothing for a drop off of film and $4 when the photographs are picked up). He also estimates that customers will arrive randomly at a mean rate of 20 per hour, although those who find the drive-through lane full will be forced to leave. Half of the customers who find the drive-through lane full wanted to drop off film and the other half wanted to pick up their photographs. The half who wanted to drop off film will take their business elsewhere instead. The other half of the customers who find the drive-through land full will not be lost because they will keep trying later until they can get in and pick up their photographs. George assumes that the time required to serve a customer will have an exponential distribution with a mean of 2 minutes.
CD S Ch 11-13 (a) Find L and the mean rate at which customers are lost when the number of car lengths of space provided is 2, 3, 4, and 5. (b) Calculate W from L for the cases considered in part (a). (c) Use the results from part (a) to calculate the decrease in the mean rate at which customers are lost when the number of car lengths of space provided is increased from 2 to 3, from 3 to 4, and from 4 to 5. Then calculate the average increase in profit per hour (excluding space rental costs) for each of these three cases. (d) Compare the increases in profit found in part (c) with the cost per hour of renting each car length of space. What conclusion do you draw about the number of car lengths of space that George should provide? E 3. The Reservations Office for Central Airlines has two agents answering incoming phone calls for flight reservations. In addition, one caller can be put on hold until one of the agents is available to take the call. If all three phone lines (both agent lines and the hold line) are busy, a potential customer gets a busy signal, in which case the call may go to another airline. The calls and attempted calls occur randomly at a mean rate of 15 per hour. The length of a telephone conversation has an exponential distribution with a mean of 4 minutes. Find the proportion of calls that fall into each of the following categories: (a) a caller will get to talk to an agent immediately, (b) the caller will be put on hold, and (c) the caller will get a busy signal. 4. At the Forrester Manufacturing Company, one repair technician has been assigned the responsibility of maintaining three machines. For each machine, the probability distribution of the running time before a breakdown is exponential, with a mean of 9 hours. The repair time also has an exponential distribution, with a mean of 2 hours. (a) Which queueing model fits this queueing system? (b) Use this queueing model to find the probability distribution of the number of machines not running, and the mean of this distribution. (c) Use this mean to calculate the expected time between a machine breakdown and the completion of the repair of that machine. (d) What is the expected fraction of time that the repair technician will be busy? (e) As a crude approximation, assume that the calling population is infinite and that machine breakdowns occur randomly at a mean rate of 3 every 9 hours. Compare the result from part (b) with that obtained by making this approximation while using (i) the corresponding infinite queue model and (ii) the corresponding finite queue model. (f) Repeat parts (b) and (d) when a second repair technician is made available to repair a second machine whenever more than one of these three machines require repair. (When repeating part (d), assume that the two repair technicians share the workload equally so that each one will be busy the same expected fraction of time, and then find this fraction.) 5. The Dolomite Corporation is making plans for a new factory. One department has been allocated 12 semi-automatic machines. A small number (yet to be determined) of operators will be hired to provide the machines the needed occasional servicing (loading, unloading, adjusting, setup, and so on). A decision now needs to be made on how to
CD S Ch 11-14 organize the operators to do this. Alternative 1 is to assign each operator to her own machines. Alternative 2 is to pool the operators so that any idle operator can take the next machine needing servicing. Alternative 3 is to combine the operators into a single crew that will work together on any machine needing servicing. The running time (time between completing service and the machines requiring service again) of each machine is expected to have an exponential distribution, with a mean of 150 minutes. The service time is assumed to have an exponential distribution, with a mean of 15 minutes (for Alternatives 1 and 2) or 15 minutes divided by the number of operators in the crew (for Alternative 3). For the department to achieve the required production rate, the machines must be running at least 89 percent of the time on average. (a) For Alternative 1, what is the maximum number of machines that can be assigned to an operator while still achieving the required production rate? What is the resulting utilization of each operator? (b) For Alternative 2, what is the minimum number of operators needed to achieve the required production rate? What is the resulting utilization of the operators? (c) For Alternative 3, what is the minimum size of the crew needed to achieve the required production rate? What is the resulting utilization of the crew? 6. Antonio runs a shoe repair store by himself. Customers arrive randomly to bring a pair of shoes to be repaired at a mean rate of one per hour. The time Antonio requires to repair each individual shoe has an exponential distribution with a mean of 15 minutes. According to statistical theory, this implies that the time required to repair a pair of shoes has an Erlang distribution with a mean of 30 minutes and a shape parameter of k = 2. a. Calculate the average number of pairs of shoes in the shop. b. Calculate the average amount of time from when a customer drops off a pair of shoes until they are repaired and ready to be picked up. E c. Use the Excel template for the M/G/1 model to check your answers in parts a and b. E d. Repeat part c with the Excel template for the M/Ek/1 model. 7. The maintenance base for Friendly Skies Airline has facilities for overhauling only one airplane engine at a time. Therefore, to return the airplanes to use as soon as possible, the policy has been to stagger the overhauling of the four engines of each airplane. In other words, only one engine is overhauled each time an airplane comes into the shop. Under this policy, airplanes have arrived randomly at a mean rate of one per day. The time required for an engine overhaul (once work has begun) has an exponential distribution with a mean of 12 day. A proposal has been made to change the policy so that all four engines are overhauled consecutively each time an airplane comes into the shop. This would mean that each plane would need to come to the maintenance base only one-fourth as often. Since the time required to overhaul one engine has an exponential distribution, statistical theory indicates that the time required to overhaul four engines has an Erlang distribution with a mean four times as large and with a shape parameter of k = 4. Management now needs to decide whether to continue the status quo or adopt the proposal. The objective is to minimize the average amount of flying time lost by the entire fleet per day due to engine overhauls.
CD S Ch 11-15 E a. Compare the two alternatives with respect to the average amount of flying time lost by an airplane each time it comes to the maintenance base. E b. Compare the two alternatives with respect to the average number of airplanes losing flying time due to being at the maintenance base. c. Which of these two comparisons is the appropriate one for making managements decision? Explain. The McAllister Company factory currently has two tool cribs, each with a single clerk, in its manufacturing area. One tool crib handles only the tools for the heavy machinery; the second one handles all other tools. However, for each crib, the mechanics arrive randomly to obtain tools at a mean rate of 24 per hour, and the expected service time is two minutes. Because of complaints that the mechanics coming to the tool cribs have to wait too long, it has been proposed that the two tool cribs be combined so that either clerk can handle either kind of tool as the demand arises. It is believed that the mean arrival rate to the combined two-clerk tool crib would double to 48 per hour and that the expected service time would continue to be two minutes. However, information is not available on the form of the probability distribution for service times, so it is not clear which queueing model would be most appropriate. E a. Compare the status quo and the proposal with respect to the total expected number of mechanics at the tool crib(s) and the expected waiting time (including service) for each mechanic. Do this by tabulating these data for the four queueing models where the distribution of service times is (1) exponential, (2) Erlang with shape parameter k = 2, (3) Erlang with shape parameter k = 8, and (4) degenerate (constant service times). Use available Excel templates, as well as Figures 11.10 and 4, as needed. b. Given these results, which alternative should be chosen? c. Which of the four insights presented in Section 11.8 is illustrated by these results? The car rental company, Try Harder, has been subcontracting for the maintenance of its cars in St. Louis. However, due to long delays in getting its cars back, the company has decided to open its own maintenance shop to do this work more quickly. This shop will operate 42 hours per week. Alternative 1 is to hire two mechanics (at a cost of $1,500 per week each), so that two cars can be worked on at a time. The time required by a mechanic to service a car has an Erlang distribution with a mean of five hours and a shape parameter of k = 8. Alternative 2 is to hire just one mechanic (for $1,500 per week) but to provide some additional special equipment (at a capitalized cost of $1,250 per week) to speed up the work. In this case, the maintenance work on each car is done in two stages, where the time required for each stage has an Erlang distribution with the shape parameter k = 4, where the mean is two hours for the first stage and one hour for the second stage. This implies that the probability distribution of the total time for both stages has a mean of three hours and a variance equal to the sum of the variances of the times for the individual stages. However, the form of this distribution is unknown. (It is not Erlang.) For both alternatives, the cars arrive randomly at a mean rate of 0.3 cars per hour (during work hours). The company estimates that its net lost revenue due to having its cars unavailable for rental is $150 per week per car. a. Use Figure 4 to estimate L, Lq, W, and Wq for Alternative 1. E b. Find these same measures of performance for Alternative 2.
8.
9.
CD S Ch 11-16 c. Determine and compare the expected total cost per week for these alternatives.