Queuing Analysis: Chapter Outline
Queuing Analysis: Chapter Outline
SUPPLEMENTARY
CHAPTER B Queuing Analysis
Learning Objectives
• To understand the key elements and underlying mathematical
concepts of analytical queuing models: arrival distribution, service-
time distribution, queue discipline, and queuing behavior.
• To understand the single-server queuing model and be able to
calculate and interpret the operating characteristics associated with
the model.
• To be able to apply the operating characteristic formulas for a
multiple-server queuing model to evaluate performance of practical
queuing systems.
• To understand economic trade-offs associated with designing and
managing queuing systems.
• To appreciate the importance of understanding the psychology of
waiting in designing and managing queuing systems in practical
business situations.
B2
• Amtrak, the U.S. passenger train service, pays freight railroads a fee to use
their tracks. With freight train business growing, Amtrak trains were on time
only 63 percent of the time in July, 2004. When freight trains back up, there
is a spillover effect on passenger trains. Amtrak’s Sunset Limited passenger
train between Orlando, Florida and Los Angeles hasn’t been on time once in
the previous five months. Amtrak terminated one trip in El Paso, Texas when
the Sunset Limited arrived 35 hours late—it forced Amtrak to bus and fly all
passengers between El Paso and points West. Amtrak passenger traffic is
up 6%, in part because more airline travelers are fed up with delays at the
airports.2
The first episode highlights a growing practice of segmenting customers so that pre-
mium service is provided to a few high-value customers while many low-value cus-
tomers get less attention and organizational resources. The electric utility’s call
center assigns the same number of CSRs—six—to the top 350 customers and the
next 700 customers based on value. Many organizations would gladly see customers
Value-based queuing is a that generate marginal profits leave. Value-based queuing is a method that allows or-
method that allows ganizations to prioritize customer calls based on their long-term value to the orga-
organizations to prioritize nization. Low-profitability customers are often encouraged to serve themselves on
customer calls based on their the company’s web site rather than tie up expensive telephone representatives. Such
long-term value to the decisions are similar to the notion of segmenting high-value inventory using ABC
organization.
analysis that we discussed in Chapter 12. Do you think that this decision is good
or bad? Should all customers be treated the same and be considered as important
as any other?
The second episode illustrates the interdependency between U.S. airline, passen-
ger train, and freight train networks. Major delays in one transportation system
can quickly cascade to other types of transportation with huge cost implications.
For all three types of transportation, the management of routes, schedules, queues,
capacity, and the use of priority rules are critical to the efficiency of these inter-
dependent transportation networks.
This supplement introduces basic concepts and methods of queuing analysis
that have wide applicability in manufacturing and service organizations. We focus
only on simple models; other textbooks devoted exclusively to management science
develop more complex models.
Learning Objective
To understand the key elements
ANALYZING QUEUES USING ANALYTICAL MODELS
and underlying mathematical
Many analytical queuing models exist, each based on unique assumptions about
concepts of analytical queuing
models: arrival distribution, the nature of arrivals, service times, and other aspects of the system. Some of the
service-time distribution, queue common models are
discipline, and queuing behavior.
1. Single- or multiple-channel with Poisson arrivals and exponential service times.
(This is the most elementary situation.)
Supplementary Chapter B: Queuing Analysis B3
2. Single-channel with Poisson arrivals and arbitrary service times. (Service times
may follow any probability distribution, and only the average and the standard
deviation need to be known.)
3. Single-channel with Poisson arrivals and deterministic service times. (Service
times are assumed to be constant.)
4. Single- or multiple-channel with Poisson arrivals, arbitrary service times, and no
waiting line. (Waiting is not permitted. If the server is busy when a unit arrives,
the unit must leave the system but may try to reenter at a later time.)
5. Single- or multiple-channel with Poisson arrivals, exponential service times, and
a finite calling population. (A finite population of units is permitted to arrive
for service.)
We illustrate the development of the basic queuing model for the problem of
designing an automated check-in kiosk for passengers at an airport. Suppose that
process design and facility-layout activities are currently being conducted for a new
terminal at a major airport. One particular concern is the design and layout of the
passenger check-in system. Most major airlines now use automated kiosks to speed
up the process of obtaining a boarding pass with an electronic ticket. Passengers
either enter a confirmation number or scan their electronic ticket to print a board-
ing pass. A queuing analysis of the system will help to determine if the system will
provide adequate service to the airport passengers. To develop a queuing model,
we must identify some important characteristics of the system: (1) the arrival dis-
tribution of the passengers, (2) the service-time distribution for the check-in oper-
ation, and (3) the waiting-line, or queue discipline for the passengers.
Arrival Distribution
Defining the arrival distribution for a waiting line consists of determining how many
customers arrive for service in given periods of time, for example, the number of
passengers arriving at the check-in kiosk during each 1-, 10-, or 60-minute period.
Since the number of passengers arriving each minute is not a constant, we need to
define a probability distribution that will describe the passenger arrivals. The choice
of time period is arbitrary—as long as the same time period is used consistently—
and is often determined based on the rate of arrivals and the ease by which the
data can be collected. Generally, the slower the rate of arrivals, the longer the time
period chosen.
For many waiting lines, the arrivals occurring in a given period of time appear
to have a random pattern—that is, although we may have a good estimate of the
total number of expected arrivals, each arrival is independent of other arrivals, and
we cannot predict when it will occur. In such cases, a good description of the
arrival pattern is obtained from the Poisson probability distribution:
xe
P(x) for x 0, 1, 2, . . . (B.1)
x!
where
x number of arrivals in a specific period of time
average, or expected, number of arrivals for the specific period of time
e 2.71828
For the passenger check-in process, the wide variety of flight schedules and the
variation in passenger arrivals for the various flights cause the number of passen-
gers arriving to vary substantially. For example, data collected from the actual oper-
ation of similar facilities show that in some instances, 20 to 25 passengers arrive
during a 10-minute period. At other times, however, passenger arrivals drop to
three or fewer passengers during a 10-minute period. Because passenger arrivals
B4 Supplementary Chapter B: Queuing Analysis
Exhibit B.1
Poisson Distribution of
Passenger Arrivals
Supplementary Chapter B: Queuing Analysis B5
Service-Time Distribution
A service-time probability distribution is needed to describe how long it takes to
check in a passenger at the kiosk. This length of time is referred to as the service
time for the passenger. Although many passengers will complete the check-in process
in a relatively short time, others might take a longer time because of unfamiliarity
with the kiosk operation, ticketing problems, flight changes, and so on. Thus we
expect service times to vary from passenger to passenger. In the development of
waiting-line models, operations researchers have found that the exponential prob-
ability distribution can often be used to describe the service-time distribution. Equa-
tion (B.2) defines the exponential probability distribution
ƒ(t) et for t 0 (B.2)
where
t service time (expressed in number of time periods)
average or expected number of units that the service facility can handle
in a specific period of time
e 2.71828
It is important to use the same time period used for defining arrivals in defining
the average service rate! If we use an exponential service-time distribution, the prob-
ability of a service being completed within t time periods is given by
P(Service time t Time periods) 1 et (B.3)
By collecting data on service times for similar check-in systems in operation at other
airports, we find that the system can handle an average of 10 passengers per 10-
minute period. Using a mean service rate of 10 customers per 10-minute period
in Equation (B.3), we find that the probability of a check-in service being completed
within t 10-minute periods is
P(Service time t 10-minute time periods) 1 e10t
Now we can compute the probability that a passenger completes the service within
any specified time, t. For example, for 1 minute, we set t 0.1 (as a fraction of a
10-minute period). Some example calculations are
P(Service time 1 minute) 1 e10(0.1) 1 e1 .6321
P(Service time 2.5 minutes) 1 e10(0.25) 1 e2.5 .9179
Thus, using the exponential distribution, we would expect 63.21 percent of the
passengers to be serviced in 1 minute or less, and 91.79 percent in 21/2 minutes or
less. Exhibit B.2 shows graphically the probability that t minutes or less will be
required to service a passenger.
In the analysis of a specific waiting line, we want to collect data on actual service
times to see if the exponential distribution assumption is appropriate. If you find other
service-time patterns (such as a normal service-time probability distribution or a con-
stant service-time distribution), the exponential distribution should not be used.
Queue Discipline
A queue discipline is the manner in which new arrivals are ordered or prioritized for A queue discipline is the
service. For the airport problem, and in general for most customer-oriented wait- manner in which new arrivals
ing lines, the waiting units are ordered on a first-come, first-served (FCFS) basis— are ordered or prioritized for
referred to as an FCFS queue discipline. Other types of queue disciplines are also service.
prevalent. These include
• Shortest processing time (SPT), which we discussed in Chapter 14. SPT tries to
maximize the number of units processed, but units with long processing times
must wait long periods of time to be processed, if at all.
B6 Supplementary Chapter B: Queuing Analysis
Exhibit B.2
Probability that a Passenger Will
be Serviced in t Minutes
Learning Objective
To understand the single-server
SINGLE-SERVER QUEUING MODEL
queuing model and be able to
The queuing model presented in this section can be applied to waiting-line situa-
calculate and interpret the
operating characteristics tions that meet these assumptions or conditions:
associated with the model. 1. The waiting line has a single server.
2. The pattern of arrivals follows a Poisson probability distribution.
3. The service times follow an exponential probability distribution.
4. The queue discipline is first-come, first-served (FCFS).
5. No balking or reneging.
Supplementary Chapter B: Queuing Analysis B7
Because we have assumed that these conditions are applicable to the airport
check-in problem, we can use this queuing model to analyze the operation. We have
already concluded that the mean arrival rate is 9 passengers per 10-minute
period and that the mean service rate is 10 passengers per 10-minute period.
Using the assumptions of Poisson arrivals and exponential service times, quantita-
tive analysts have developed the following expressions to define the operating char-
acteristics of a single-channel waiting line:
1. the probability that the service facility is idle (that is, the probability of 0 units
in the system):
P0 (1 /) (B.4)
2. the probability of n units in the system:
Pn (/)nP0 (B.5)
3. the average number of units waiting for service:
2
Lq (B.6)
( )
4. the average number of units in the system:
L Lq / (B.7)
5. the average time a unit spends waiting for service:
Wq Lq/ (B.8)
6. the average time a unit spends in the system (waiting time plus service time):
W Wq 1/ (B.9)
7. the probability that an arriving unit has to wait for service:
Pw / (B.10)
The values of the mean arrival rate, , and the mean service rate, , are clearly
important components in these formulas. From Equation (B.10), we see that the
ratio of these two values, /, is simply the probability that an arriving unit must
wait because the server is busy. Thus, / is often referred to as the utilization
factor for the waiting line. The formulas for determining the operating character-
istics of a single-server waiting line presented in Equations (B.4) through (B.10)
are applicable only when the utilization factor, /, is less than 1. This condition
occurs when the mean service rate, , is greater than the mean arrival rate, , and
hence when the service rate is sufficient to process or service all arrivals.
Returning to the airport check-in problem, we see that with 9 and 10,
we can use Equations (B.4) through (B.10) to determine the operating characteris-
tics of the screening operation. This is done as follows:
P0 (1 /) (1 9/10) .10
2 92
Lq 81\10 8.1 passengers
( ) 10(10 9)
L Lq / 8.1 9/10 9.0 passengers
Wq Lq/ 8.1/9 .9 (Note that this refers to the number of 10-minute
periods, or equivalently, 9 minutes per passenger)
W Wq 1/ 0.9 hour 1/10 hour one 10-minute period, or
equivalently, 10 minutes per passenger
Pw / 9/10 .90
B8 Supplementary Chapter B: Queuing Analysis
Using this information, we can learn several important things about the check-
in operation. In particular, we see that passengers wait an average of 9 minutes at
the kiosk. With this as the average, many passengers wait even longer. In airport
operations with passengers rushing to meet plane connections, this waiting time
might be judged to be undesirably high. In addition, the facts that the average
number of passengers waiting in line is 8.1 and that 90 percent of the arriving
passengers must wait to check in might suggest to the operations manager that
something should be done to improve the efficiency of the process.
These operating characteristics are based on the assumption of an arrival rate
of 9 and a service rate of 10 per 10-minute period. As the figures are based on
airport planners’ estimates, they are subject to forecasting errors. It is easy to
examine the effects of a variety of assumptions about arrival and service rates on
the operating characteristics by using a spreadsheet such as the one in Exhibit
B.3. In this spreadsheet, we compute the operating characteristics for the baseline
assumptions and also examine the effect of changes in the mean arrival rate from
7 to 10 passengers per period.
What do the data in this figure tell us about the design of this process? They
tell us that if the mean arrival rate is 7 passengers per period, the system functions
acceptably. On the average, only 1.63 passengers are waiting and the average wait-
ing time of 0.23(10 minutes) 2.3 minutes appears acceptable. However, we see
that the mean arrival rate of 9 passengers per period provides undesirable waiting
characteristics, and if the rate increases to 10 passengers per period, the system as
proposed is completely inadequate. When , the operating characteristics are
not defined, meaning that these times and numbers of passengers grow infinitely
large (that is, when → ). These results show that airport planners need to
consider design modifications that will improve the efficiency of the check-in
process.
If a new process can be designed that will improve the passenger-service rate,
Equations (B.4) through (B.10) can be used to predict operating characteristics
under any revised mean service rate, . Developing a spreadsheet with alternative
mean service rates provides the information to determine which, if any, of the screen-
ing facility designs can handle the passenger volume acceptably.
Computing the probability of more or less than x units arriving requires us to
use Equation (B.5) and the following two equations:
P(Number of arrivals
x) 1 P(Number of arrivals x) (B.11)
P(Number of arrivals x) 1 P(Number of arrivals x) (B.12)
Exhibit B.3
Spreadsheet for Single-Server
Queuing Model
(Single Server Queue.xls)
Supplementary Chapter B: Queuing Analysis B9
These equations are used to simplify the calculations. For example, to find the prob-
ability that more than 4 customers are waiting for service, we would need to sum the
probabilities associated with 5, 6, 7, . . . up to a (theoretically) infinite number.
From Equation (B.11), we simply need to find the probabilities associated with 4
or less, sum them up, and subtract them from 1.0. With 9 and 10 and
using Equation (B.5), we have
Therefore, the probability that more than 4 customers would be waiting for service
is 1 .4095 .5905.
Exhibit B.4
A Two-Server Queuing System
B10 Supplementary Chapter B: Queuing Analysis
2. The arrivals follow a Poisson probability distribution with a mean arrival rate
of .
3. The service times have an exponential distribution.
4. The mean service rate, , is the same for each server.
5. The arrivals wait in a single line and then move to the first open server for
service.
6. The queue discipline is first-come, first-served (FCFS).
7. No balking or reneging is allowed.
Using these assumptions, operations researchers have developed formulas for
determining the operating characteristics of the multiple-server waiting line. Let
k number of channels
mean arrival rate for the system
mean service rate for each channel
The following equations apply to multiple-server waiting lines for which the over-
all mean service rate, k, is greater than the mean arrival rate, . In such cases, the
service rate is sufficient to process all arrivals.
1. the probability that all k service channels are idle (that is, the probability of
zero units in the system):
1
P0
k1 (/)n (/)k
n !
(k 1)! k (B.13)
n0
Exhibit B.5
Number of Servers (k) Values of P0 for Multiple-Server
Queuing Model
Ratio / 2 3 4 5
0.15 0.8605 0.8607 0.8607 0.8607
0.20 0.8182 0.8187 0.8187 0.8187
0.25 0.7778 0.7788 0.7788 0.7788
0.30 0.7391 0.7407 0.7408 0.7408
0.35 0.7021 0.7046 0.7047 0.7047
0.40 0.6667 0.6701 0.6703 0.6703
0.45 0.6327 0.6373 0.6376 0.6376
0.50 0.6000 0.6061 0.6065 0.6065
0.55 0.5686 0.5763 0.5769 0.5769
0.60 0.5385 0.5479 0.5487 0.5488
0.65 0.5094 0.5209 0.5219 0.5220
0.70 0.4815 0.4952 0.4965 0.4966
0.75 0.4545 0.4706 0.4722 0.4724
0.80 0.4286 0.4472 0.4491 0.4493
0.85 0.4035 0.4248 0.4271 0.4274
0.90 0.3793 0.4035 0.4062 0.4065
0.95 0.3559 0.3831 0.3863 0.3867
1.00 0.3333 0.3636 0.3673 0.3678
1.20 0.2500 0.2941 0.3002 0.3011
1.40 0.1765 0.2360 0.2449 0.2463
1.60 0.1111 0.1872 0.1993 0.2014
1.80 0.0526 0.1460 0.1616 0.1646
2.00 0.1111 0.1304 0.1343
2.20 0.0815 0.1046 0.1094
2.40 0.0562 0.0831 0.0889
2.60 0.0345 0.0651 0.0721
2.80 0.0160 0.0521 0.0581
3.00 0.0377 0.0466
3.20 0.0273 0.0372
3.40 0.0186 0.0293
3.60 0.0113 0.0228
3.80 0.0051 0.0174
4.00 0.0130
4.20 0.0093
4.40 0.0063
4.60 0.0038
4.80 0.0017
1
W 0.026 0.126 multiples of 10-minute periods,
10
or 1.26 minutes/passenger
1 9 20
20 9
2
Pw (.3793) .279
2! 10
These operating characteristics suggest that the two-server operation would handle
the volume of passengers extremely well. Specifically, note that the total time in the
system is an average of only 1.26 minutes per passenger, which is excellent. The
percentage waiting is 27.9 percent, which is acceptable, especially in light of the
short average waiting time.
Exhibit B.6 is a spreadsheet designed to compute operating characteristics for up
to eight servers in the multiple-server queuing model using the arrival and service
rates for the security-screening example. (The Microsoft Excel function HLOOKUP
is used to compute the summation and the term (/)n/n! in the denominator of
P0. The function FACT computes factorials.) With three servers, we see a signifi-
cant improvement over two servers in the operating characteristics; beyond this, the
improvement is negligible. In addition, we can use the spreadsheet to show that
even if the mean arrival rate for passengers exceeds the estimated 9 passengers per
hour, the two-channel system should operate nicely.
What is the probability that less than four customers are waiting for service
when k 2, 9, and 10 passengers per time period? Using the spreadsheet,
we can calculate the probabilities when x 0 as .3793, x 1 as .3414, x 2 as
.1536, x 3 as .0691, and x 4 as .0311. Using Equation (B.13), we sum the
probabilities from P0 to P4 and then subtract from 1 to arrive at the probability
of four or more customers waiting for service at 1 .9745 .0255. Adding the
second server greatly improves system performance, as the results show.
Exhibit B.6 Spreadsheet for Multiple-Server Queuing Model (Multiple Server Queue.xls)
Supplementary Chapter B: Queuing Analysis B13
Exhibit B.7
System k System Cost L Passenger Cost Total Cost Economic Analysis of Check-In
System Design
Single-server 1 0.167(1) 0.167 9 0.83(9) 7.47 $7.64
Two-server 2 0.167(2) 0.334 1.13 0.83(1.13) 0.94 $1.27
OM SPOTLIGHT
Airport Security Wait Times3
The U.S. Transportation Security Administration (TSA) set up a new
web site that provides airport-by-airport information on the average
wait times. The site provides hourly and daily average wait times
based on last month’s data. A longer-term goal is to provide real-
time hourly updates. In 2004, the longest (maximum) time waiting in line to get to the
metal detector was 36 minutes at a major U.S. airport. Average waiting times range from
a few minutes to 30 minutes. For example, the TSA recorded the wait at the main secu-
rity checkpoint at Hartsfield International Airport in Atlanta at 7 A.M. on Monday, August
9, 2004 averaged 26 minutes. Monday morning is a peak time for most airports. The data
are collected by security screeners who give passengers a card with their arrival time on
it, which is collected when they get to the metal detector.
B14 Supplementary Chapter B: Queuing Analysis
Supermarkets place “impulse” items such as candy, batteries, and other small items
as well as magazines near checkouts to grab customers’ attention. The Postal Service
has been experimenting with video displays that not only distract customers but
also inform them of postal procedures so that they can speed up their transactions.
Technology is alleviating queuing in many service industries today. For example,
rental car firms use automatic tellers for fast check in and out and are working
on radio frequency technology to entirely skip waiting in lines to get or return a
vehicle. Airlines allow their customers to print out boarding passes at airport kiosks
or on their own printer to speed the check-in process. Thus, queuing in operations
management entails much more than some analytical calculations and requires good
management skills.
SOLVED PROBLEMS
SOLVED PROBLEM #1
The reference desk of a large library receives requests c. L Lq / 4.1667 10/12
for assistance at a mean rate of 10 requests per hour, 5.000 requests (Equation B.7)
and it is assumed that the desk has a mean service rate
d. Wq Lq/ 4.1667/10
of 12 requests per hour.
0.4167 hour (Equation B.8)
a. What is the probability that the reference desk is
W Wq 1/
idle?
0.4167 hour 1/12 hour
b. What is the average number of requests that will 0.5 hour, or 30 minutes (Equation B.9)
be waiting for service?
e. Pw / 10/12 .8333 (Equation B.10)
c. What is the average number of requests in the sys-
f. Use Equation (B.5) to compute the following:
tem?
d. What is the average waiting time plus service time P0 (1 /) (1 10/12) .1667
for a request for assistance? P1 (/)nP0 (10/12)1(.1667) .1389
P2 (/)nP0 (10/12)2(.1667) .1157
e. What is the utilization factor? P3 (/)nP0 (10/12)3(.1667) .0965
f. What is the probability of more than three requests? Sum of P( x) .5178
SOLVED PROBLEM #2
A fast-food franchise operates a drive-up window. Or- cars in front have been served. By hiring a second per-
ders are placed at an intercom station at the back of the son to help take and fill orders, the manager hopes to
parking lot. After placing an order, the customer pulls improve service. With one person filling orders, the av-
up and waits in line at the drive-up window until the erage service time for a drive-up customer is 2 minutes;
B16 Supplementary Chapter B: Queuing Analysis
1. What are six typical performance measures for 4. Define a “queue discipline” and why this is impor-
waiting-line (queuing) models? tant?
2. What information is necessary to analyze a waiting 5. Explain how resource utilization is measured in
line? single- and multiple-server queuing models.
3. What probability distributions are used in basic 6. Are the performance relationships in queuing models
queuing models? Explain. linear in nature? Why or why not? Explain.
Supplementary Chapter B: Queuing Analysis B17
7. List the assumptions underlying the basic single- 10. Does customers’ value over their buying life figure
channel waiting-line model. into basic queuing model logic? Explain.
8. What are the assumptions of the multiple-channel 11. How do restaurants handle the psychology of wait-
waiting-line model? ing as customers arrive for dinner? Explain.
9. Can queuing models predict the size of the queue 12. Explain a waiting-line multiple-server situation where
(waiting line) when vehicle traffic is limited to two you jockeyed for position and it improved or hin-
lanes instead of three lanes? What if traffic is re- dered your service. Did other customers get mad?
duced to one lane? Make up a simple problem and Explain.
explain what you discover.
The following waiting-line problems are all based on b. What is the average number of trucks in the
the assumptions of Poisson arrivals and exponential queue?
service times. c. What is the average number of trucks in the
1. Arrivals to a single-server queue occur at an aver- system?
age rate of 3 per minute. Develop the probability d. What is the average time a truck spends in the
distribution for the number of arrivals for x 0 queue waiting for service?
through 10. e. What is the average time a truck spends in the
system?
2. Trucks using a single-server loading dock have a f. What is the probability that an arriving truck
mean arrival rate of 12 per day. The loading/ will have to wait?
unloading rate is 18 per day. g. What is the probability that more than two
a. What is the probability that the truck dock will trucks are waiting for service?
be idle? 5. Marty’s Barber Shop has one barber. Customers
b. What is the average number of trucks waiting arrive at a rate of 2.2 per hour, and haircuts are
for service? given at an average rate of 5 customers per hour.
c. What is the average time a truck waits for the
loading or unloading service? a. What is the probability that the barber is idle?
d. What is the probability that a new arrival will b. What is the probability that one customer is
have to wait? receiving a haircut and no one is waiting?
e. What is the probability that more than three c. What is the probability that one customer is
trucks are waiting for service? receiving a haircut and one customer is waiting?
d. What is the probability that one customer is
3. A mail-order nursery specializes in European beech receiving a haircut and two customers are wait-
trees. New orders, which are processed by a single ing?
shipping clerk, have a mean arrival rate of six per e. What is the probability that more than two cus-
day and a mean service rate of eight per day. tomers are waiting?
a. What is the average time an order spends in the f. What is the average time a customer waits for
queue waiting for the clerk to begin service? service?
b. What is the average time an order spends in the 6. Trosper Tire Company has decided to hire a new
system? mechanic to handle all tire changes for customers
4. Assume trucks arriving for loading/unloading at a ordering new tires. Two mechanics are available
truck dock form a single-server waiting line. The for the job. One mechanic has limited experience
mean arrival rate is four trucks per hour and the and can be hired for $7 per hour. It is expected that
mean service rate is five trucks per hour. this mechanic can service an average of three cus-
a. What is the probability that the truck dock will tomers per hour. A mechanic with several years of
be idle? experience is also being considered for the job. This
B18 Supplementary Chapter B: Queuing Analysis
mechanic can service an average of four customers b. If you were to drive up to the facility, how many
per hour, but must be paid $10 per hour. Assume cars would you expect to see waiting and being
that customers arrive at the Trosper garage at the serviced?
rate of two per hour. c. What is the average time waiting for service?
a. Compute waiting-line operating characteristics d. What is the probability an arriving car will have
for each mechanic. to wait?
b. If the company assigns a customer-waiting cost e. What is the probability that more than four ve-
of $15 per hour, which mechanic provides the hicles are waiting for service?
lower operating cost? f. As a potential customer of the system, would
you be satisfied with these waiting-line charac-
7. Agan Interior Design provides home and office dec- teristics? How do you think managers could go
orating assistance. In normal operation an average about assessing its customers’ feelings about the
of 2.5 customers arrive per hour. One design con- current system?
sultant is available to answer customer questions
and make product recommendations. The consul- 11. To improve its customer service, Keuka Park Sav-
tant averages 10 minutes with each customer. ings and Loan (Problem 10) wants to investigate the
effect of a second drive-in teller window. Assume a
a. Compute operating characteristics for the cus- mean arrival rate of 10 cars per hour. In addition,
tomer waiting line. assume a mean service rate of 12 cars per hour for
b. Service goals dictate that an arriving customer each window. What effect would adding a new teller
should not wait for service more than an aver- window have on the system? Does this system ap-
age of 5 minutes. Is this goal being met? What pear acceptable?
action do you recommend?
c. If the consultant can reduce the average time 12. Fore and Aft Marina is a new marina planned for a
spent with customers to 8 minutes, will the ser- location on the Ohio River near Madison, Indiana.
vice goal be met? Assume that Fore and Aft decides to build one dock-
ing facility and expects a mean arrival rate of 5 boats
8. Pete’s Market is a small local grocery store with one per hour and a mean service rate of 10 boats per
checkout counter. Shoppers arrive at the checkout hour.
lane at an average rate of 15 customers per hour
and the average order takes 3 minutes to ring up a. What is the probability that the boat dock will
and bag. What information would you develop to be idle?
help Pete analyze the current operation? If Pete b. What is the average number of boats that will
does not want the average time waiting for service be waiting for service?
to exceed 5 minutes, what would you tell him about c. What is the average time a boat will wait for
the current system? service?
d. What is the average time a boat will spend at
9. Refer to Problem 8. After reviewing the analysis, the dock?
Pete felt it would be desirable to hire a full-time per- e. What is the probability that more than 2 boats
son to assist in the checkout operation. Pete believed are waiting for service?
that if this employee assisted the cashier, average f. If you were the owner, would you be satisfied
service time could be reduced to 2 minutes. How- with this level of service?
ever, Pete was also considering installing a second
checkout lane, which could be operated by the new 13. The owner of the Fore and Aft Marina in Problem
person. This would provide a two-server system 12 is investigating the possibility of adding a second
with the average service time of 3 minutes for each dock. Assume a mean arrival rate of 5 boats per
customer. Should Pete use the new employee to as- hour for the marina and a mean service rate of 10
sist on the current checkout counter or to operate a boats per hour for each server.
second counter? Justify your recommendation. a. What is the probability that the boat dock will
10. Keuka Park Savings and Loan currently has one be idle?
drive-in teller window. Cars arrive at a mean rate b. What is the average number of boats that will
of 10 per hour. The mean service rate is 12 cars per be waiting for service?
hour. c. What is the average time a boat will wait for
service?
a. What is the probability that the service facility d. What is the average time a boat will spend at
will be idle? the dock?
Supplementary Chapter B: Queuing Analysis B19
e. If you were the owner, would you be satisfied c. As a customer of Big Al’s, do you think the sys-
with this level of service? tem favors the customer? If you were Al, what
14. The City Beverage Drive-Thru is considering a two- would be your attitude toward this service level?
server system. Cars arrive at the store at the mean 18. Refer to the Agan Interior Design situation in Prob-
rate of 6 per hour. The service rate for each server lem 7. Agan is evaluating two alternatives:
is 10 per hour. 1. use one consultant with an average service time
a. What is the probability that both servers are idle? of 8 minutes per customer;
b. What is the average number of cars waiting for 2. expand to two consultants, each of whom has an
service? average service time of 10 minutes per customer.
c. What is the average time waiting for service? If the consultants are paid $16 per hour and the cus-
d. What is the average time in the system? tomer waiting time is valued at $25 per hour, should
e. What is the probability of having to wait for Agan expand to the two-consultant system? Explain.
service?
19. Refer to Solved Problem 2. Space is available to in-
15. Consider a two-server waiting line with a mean ar- stall a second drive-up window adjacent to the first.
rival rate of 50 per hour and a mean service rate of The manager is considering adding such a window.
75 per hour for each server. One person will be assigned to service customers at
a. What is the probability that both servers are idle? each window.
b. What is the average number of cars waiting for a. Determine the average customer waiting time for
service? this two-server system.
c. What is the average time waiting for service? b. What percentage of the time will both windows
d. What is the average time in the system? be idle?
e. What is the probability of having to wait for c. What design would you recommend: one atten-
service? dant at one window, two attendants at one win-
16. For a two-server waiting line with a mean arrival dow, or two attendants and two windows with
rate of 14 per hour and a mean service rate of 10 one attendant at each window?
per hour per server, determine the probability that 20. Design a spreadsheet similar to Exhibit B.3 to study
an arrival must wait. What is the probability of wait- changes in the mean service rate from 10 to 15 for
ing if the system is expanded to three servers? 9 passengers per minute.
17. Big Al’s Quickie Carwash has two wash bays. Each 21. Using the spreadsheet in Exhibit B.6 (Multiple-
bay can wash 15 cars per hour. Cars arrive at the Server Queue.xls), determine the effect of increasing
carwash at the rate of 15 cars per hour on the aver- passenger arrival rates of 10, 12, 14, 16, and 18 on
age, join the waiting line, and move to the next open the operating characteristics of the airport security
bay when it becomes available. screening example.
a. What is the average time waiting for a bay?
b. What is the probability that a customer will have
to wait?
CASES
officers and sheriffs, building permit inspectors, and the lem with recommendations so he could take it to the
dog warden—all trying, he thought, to see what he was Bourbon County Board of Supervisors for their ap-
copying. The line for the copying machine often ex- proval. The board faced deficit spending last fiscal year,
tended out into the hallways of the courthouse. so the trade-offs between service and cost must be
Morris mentioned his frustration with the copying clearly presented to the board.
machine problem to Judge Hamlet and his summer in- Gifford’s experience with analyzing similar prob-
tern, Dot Gifford. Gifford was home for the summer lems at school helped her know what type of informa-
and working toward a joint MBA/JD degree from a lead- tion and data were needed. After several weeks of
ing university. working on this project, she developed the information
“Mr. Morris, there are ways to find out if that one contained in Exhibits B.8, B.9, and B.10.
copying machine is adequate to handle the demand. If Gifford was not quite as confident in evaluating this
you can get the judge to let me analyze the situation, I situation as others because the customer mix and asso-
think I can help out. We had a similar problem at the ciated labor costs seemed more uncertain in the county
law school with word processors and at the business courthouse. In the law school situation, only secretaries
school with student lab microcomputers.” used the word processing terminals; in the business
The next week Judge Hamlet gave Gifford the go- school situation, students were the ones complaining
ahead to work on the copying machine problem. He about long waiting times to get on a microcomputer
asked her to write a management report on the prob- terminal. Moreover, the professor guiding these two
Exhibit B.8
Bourbon County Court— Customer Customer Customer Customer Customer
Customer Arrivals per Hour* Arrivals in Arrivals in Arrivals in Arrivals in Arrivals in
One Hour One Hour One Hour One Hour One Hour
1 5 11 10 21 3 31 11 41 14
2 9 12 17 22 9 32 8 42 7
3 7 13 18 23 11 33 9 43 4
4 13 14 14 24 10 34 8 44 7
5 7 15 11 25 12 35 6 45 7
6 7 16 16 26 4 36 8 46 2
7 7 17 5 27 8 37 14 47 4
8 11 18 6 28 9 38 12 48 7
9 8 19 8 29 9 39 11 49 2
10 6 20 13 30 9 40 15 50 8
*A sample of customer arrivals at the copying machine was taken for five consecutive 9-hour work days plus 5
hours on Saturday for a total of 50 observations. The mean arrival rate is 8.92 arrivals per hour.
Exhibit B.9
Bourbon County Court— Obs. Hours Obs. Hours Obs. Hours Obs. Hours Obs. Hours
Copying Service Times* No. per Job No. per Job No. per Job No. per Job No. per Job
1 0.0700 11 0.1253 21 0.1754 31 0.0752 41 0.2005
2 0.1253 12 0.1754 22 0.0700 32 0.1002 42 0.0501
3 0.0752 13 0.0301 23 0.1253 33 0.0250 43 0.0150
4 0.2508 14 0.1002 24 0.0752 34 0.0752 44 0.0501
5 0.0226 15 0.0752 25 0.2508 35 0.0501 45 0.0527
6 0.1504 16 0.3009 26 0.0752 36 0.0301 46 0.1203
7 0.0501 17 0.0752 27 0.0752 37 0.0752 47 0.1253
8 0.0250 18 0.0376 28 0.1002 38 0.0501 48 0.1053
9 0.0150 19 0.0501 29 0.0388 39 0.0075 49 0.1253
10 0.2005 20 0.0226 30 0.0978 40 0.0602 50 0.0301
*A sample of customers served at the copying machine was taken for five consecutive 9-hour work days plus 5
hours on Saturday for a total of 50 observations. The average service time is 0.0917 hour per copying job, or 5.499
minutes per job. The equivalent service rate is 10.91 jobs per hour (that is, 10.91 jobs/hour (60 minutes/hour)/
5.5 minutes/job).
Supplementary Chapter B: Queuing Analysis B21
Exhibit B.10
Mix of Customers Cost of Average Bourbon County Court—Cost
Resource Category in Line (%) Direct Wages per Hour and Customer Mix*
Lease and maintenance cost of copying na $18,600
machine per year @ 250 days/year
Average hourly copier variable cost na $5/hour
(electric, ink, paper, etc.)
Secretaries 50% $18.75
Clerks of the Court 20% $22.50
Building Inspectors and Dog Warden 10% $28.40
Police Officers and Sheriffs 10% $30.80
Attorneys 10% $100.00
*The mix of customers standing in line was collected at the same time as the data in Exhibits B.8 and B.9. Direct
wages do include employee benefits, but not work opportunity costs, ill-will costs, etc.
past school projects had suggested using queuing mod- 2. What are the economics of the situation using queu-
els for one project and simulation for the other project. ing model analysis?
Gifford was never clear on how the method of analysis 3. What are your final recommendations using queu-
was chosen. Now, she wondered which methodology ing model analysis?
she should use for the Bourbon County Court situation.
To organize her thinking Gifford listed a few of the 4. Advanced Assignment (requires the use of Crystal
questions she needed to address as follows: Ball on the CD-ROM). Do the customer arrival and
service empirical (actual) distributions in the case
1. Assuming a Poisson arrival distribution and an ex- match the theoretical distributions assumed in queu-
ponential service time distribution, apply queuing ing models?
models to the case situation and evaluate the results.
ENDNOTES
1 Brady, D., “Why Service Stinks,” Business Week October 23, 2000, pp. 118–128. This episode is partially based on this article.
2 Machalaba, D., “Taking the Slow Train: Amtrak Delays Rise Sharply,” Wall Street Journal August 10, 2004, pp. D1–D2.
3 Schatz, A., “Airport Security-Checkpoint Wait Times Go Online,” Wall Street Journal August 10, 2004, p. D2.
4 Graessel, Bob, and Zeidler, Pete, “Using Quality Function Deployment to Improve Customer Service,” Quality Progress November 1998,
pp. 59–63.