ISOM 2700: Operations Management

Session 10: Service Management II

Dongwook Shin
Dept. ISOM, HKUST Business School
Announcement
• Quiz 3
• Release on Oct 13 (6pm)
• Due on Oct 17 (6pm)
• Covers course material from Oct 6 to Oct 13

• Midterm exam review

• Oct 18 (Mon)
• Let me know in advance anything you want me to discuss in
the class

• No class on Oct 20 (Wed)

1
Insights So Far

• No variability in inter-arrival and service times, a queue

will form only when the arrival rate is greater than the
service rate

• When there is variability in inter-arrival and/or service

times, a queue will form even when the average arrival
rate is lower than the average service rate

2
 Recap: M/M/1 Model Utilization
𝜌 = 𝜆/𝜇

Inter-arrival
Rate 𝜆 Single
Buffer
Server

Avg. Number Waiting Avg. Number in Service

𝐿! = 𝜌" /(1 − 𝜌) 𝐿# = 𝜌 = 𝜆/𝜇

Avg. Waiting time Avg. Service time

𝑊! = 𝐿! /𝜆 𝑊# = 1/𝜇
There are n
customers in
the system
Avg. Number in the System
with probability
𝐿 = 𝜌/(1 − 𝜌)
𝑃$ = 𝜌$ (1 − 𝜌)
Avg. Time in System
𝑊 = 𝐿/𝜆 3
Recap: M/M/s Model Utilization
𝜌 = 𝜆/𝑠𝜇

Server 1
Inter-arrival
Rate 𝜆 Server 2
Buffer
⋅⋅⋅
Server s

Avg. Number Waiting Avg. Number in Service

𝐿! (using table) 𝐿# = 𝑠𝜌 = 𝜆/𝜇

Avg. Waiting time Avg. Service time

𝑊! = 𝐿! /𝜆 𝑊# = 1/𝜇

Avg. Number in the System

𝐿 = 𝐿! + 𝐿# = 𝐿! + 𝜆/𝜇
Avg. Time in System
𝑊 = 𝐿/𝜆 4
 Table for
Values of 𝐿/

5
M/M/s Queueing Spreadsheet
MMS.xlsx

6
Learning Objectives: Session 10

• Economic implications: generating a staffing plan

• Waiting time formula for non-exponential distributions

• Impact of pooling: economies of scale

• Psychology of waiting

7
Example: Emergency Room Staffing

The average arrival rate to the emergency room of Queen

Elizabeth Hospital is 5.9 patients per hour
Average time spent by a doctor with a patient is 29
minutes. That is, each doctor can serve an average of
(60/29) = 2.07 patients per hour
The Hospital Authority wants the average waiting time to
be no more than 15 minutes
What is the minimum number of doctors in the
emergency room? L0 = 𝜆𝑊/ ≤ (5.9/ℎ𝑜𝑢𝑟)(0.25 hour)
=1.475. To achieve this, 4 doctors are needed
8
Example: Emergency Room Staffing
(Extended)

A patient right group requests the hospital authority to

reduce the average waiting time to no more than 1
minute.
The Hospital Authority replies, “This means reducing the
average time by 15 times, and hence increasing the
number of doctors by 15 times.”
Is that correct?
No. Only 2 more doctors (total 6 doctors) are needed.

9
Example: Students at ASO
• Students arrive at the ASO at an average of one every
15 minutes and their requests take on average 10
minutes to be processed. The service counter is staffed
by only one clerk, Judy, who works 8 hours per day.
Assume inter-arrival and service times are exponential
a) What percentage of time is Judy idle?
b) How much time, on average, does a student spend waiting in
line?
c) How long is the (waiting) line on average?
d) What is the probability that an arriving student (just before
entering the ASO) will find at least one other student waiting
in line?

10
Example: Students at ASO
a) What percentage of time is Judy idle?
Utilization = Flow rate / Capacity = 4 per hr / 6 per hr = 0.67
Judy is idle 33% (=1-Utilization) of the time
b) How much time, on average, does a student spend
waiting in line?
𝜆 4
𝑊! = = = 0.33 ℎ𝑜𝑢𝑟 = 20 𝑚𝑖𝑛𝑢𝑡𝑒𝑠
𝜇(𝜇 − 𝜆) 6(6 − 4)

c) How long is the (waiting) line on average?

𝜆" 4"
𝐿! = = = 1.33
𝜇(𝜇 − 𝜆) 6 6 − 4

11
Example: Students at ASO
d) What is the probability that an arriving student will
find at least one other student waiting in line?

When there is at least one other student in the system, it implies at

least two will be there (one being served and one waiting). The
probability is 1 – (P0 + P1).

Compute Pn = (1 - l / µ) (l / µ) n
→ P0 = 0.3333 and P1 = 0.2222
Therefore, 1 – (P0 + P1 ) = 0.4445

The probability that an arriving student will find at least one other
student waiting in line is 0.4445

12
Example: Students at ASO (Extended)
The managers of the ASO estimate that the time a student spends
waiting in line costs them (due to goodwill loss and so on) $10 per
hour. To reduce the time a student spends waiting, they need to
improve Judy’s processing time (in Example 1). There are 2 options:
A. Install a computer system, with which Judy expects to be
able to complete a student request in 6 minutes (original 10
minutes)
B. Hire another temporary clerk, who will work at the same
rate as Judy
The computer system costs $99.5 per day, while the temporary clerk
gets paid $75 per day. Which option is preferred? Assume
exponential inter-arrival and service times

13
Example: Students at ASO (Extended)
• Option A: Maintain one staff at the counter with a new
computer system
Use M/M/1 model with l = 4 and µ = 10
Compute Lq = l2 / µ (µ - l) = 42 / [10(10 – 4)] = 0.267
Total cost = Waiting cost + Cost of installing a computer system
Total cost = (0.267)($10)(8 hours) + $99.5 = $120.83 per day
• Option B: Have two staff at the counter
Use M/M/s model with s = 2, r = l / (s*µ) = 4 / 12 = 0.333
Obtain Lq = 0.0833 from Excel M/M/s sheet
Total cost = Waiting cost + Cost of hiring a temporary clerk
Total cost = (0.0833)($10)(8 hours) + $75 = $81.7 per day

Conclusion: Option B is better 14

Learning Objectives: Session 10

• Economic implications: generating a staffing plan

• Waiting time formula for non-exponential distributions

• Impact of pooling: economies of scale

• Psychology of waiting

15
What if Inter-arrival and Service Time
Distributions Are Not Exponential?

• G/G/s queueing models

• General inter-arrival time distributions
• General service time distributions
• Closed-form formulas for performance measures, such as
𝐿! , 𝐿, 𝑊! , 𝑊, are not available for G/G/s models

Need a quick and dirty approximation!

16
Relative and Absolute Variability
• The standard deviation is an absolute measure of variability
• Two processes can have the same standard deviation but one can
seem much more variable than the other
• Below are random samples from two processes that have the same
standard deviation. The left one seems more variable

Average = 10 Stdev = 10 Average = 100 Stdev = 10

50 120
45
100
Inter-arrival time (min)

40

Inter-arrival time (min)

35
80
30
25 60
20
15 40
10
20
5
0 0
Observation Observation

17
Relative and Absolute Variability
• The previous slide plotted the processes on two different axes
• Here, the two are plotted relative to their average and with the
same axes
• Relative to their average, the one on the left is clearly more variable

Average = 10 Stdev = 10 Average = 100 Stdev = 10

5.0 5.0
4.5 4.5

Inter-arrival time /Average

Inter-arrival time /Average

4.0 4.0
3.5 3.5
3.0 3.0
2.5 2.5
2.0 2.0
1.5 1.5
1.0 1.0
0.5 0.5
0.0 0.0
Observation Observation

18
Coefficient of Variation (CV)
• Measure of variability of a random variable X
standard deviacon (X)
CVX=
mean (X)
• CV measures variability relative to the mean

1 1 2 2

0 3 12
CV = 1/3 CV = 2/12 = 1/6
19
Waiting Time Formula for a Single Server
Avg. Waiting Time = 𝑊" Avg. Service Time =𝑊!

Inter-arrival Rate 𝜆
Buffer Server

CV#" + CV$" Uklizakon

W! = × × W$
2 1−Uklizakon

Variability factor Utilization factor Service time factor

CV% and CV& are coefficients of variations for inter-arrival

and service time distributions, respectively
20
Effect of Utilization

CV#" + CV$" Uklizakon

W! = × × W$
2 1−Uklizakon

Variability factor Utilization factor Service time factor

Average
waiting time

30% 40% 50% 60% 70% 80% 90% 100%

Utilization 21
Effect of Variability

CV#" + CV$" Uklizakon

W! = × × W$
2 1−Uklizakon

Variability factor Utilization factor Service time factor

Average
waiting time
Increasing
variability

30% 40% 50% 60% 70% 80% 90% 100%

Utilization 22
Waiting Time Formula for Multiple,
Parallel Servers
Avg. Waiting Time = 𝑊" Avg. Service Time =𝑊!

Server
Inter-arrival Rate 𝜆
Buffer Server

Server

Waiting time formula for multiple (s) servers

2(s+1)−1
CV#" + CV$" Uklizakon W$
W! = × ×
2 1−Uklizakon s

23
Discussion

• Waiting time formula applies only when Utilization < 1

• If Utilization > 1, the system is unstable

• Waiting time formula is exact when s=1 and inter-arrival

and service time distributions are exponential

24
Why the Long Queues for the Women’s
Restroom?
2(s+1)−1
CV#$ + CV!$ U@liza@on W!
W" = × ×
2 1−U@liza@on s

• Women’s Restroom Equity Bill

• Passed NY City Counsel May 2005
• Requires women’s restrooms to
have twice the flushing capacity of
men’s restrooms
• In the context of the waiting time
equation, this law requires s to
increase for women, which
decreases the service time factor
(Wp /𝑠) and the utilization factor

• Other solutions?

25
Learning Objectives: Session 10

• Economic implications: generating a staffing plan

• Waiting time formula for non-exponential distributions

• Impact of pooling: economies of scale

• Psychology of waiting

26
Example: Pooling

Alternative 1: Common Queue Alternative 2: Separate Queues

• Assumptions
• Arrival rate of customer = 4.73 customers / min
• Service rate of each server = 1.77 customers / min
• 𝐶𝑉% = 𝐶𝑉# = 1

27
Example: Pooling
• Alternative 1: One common queue

Utilization = (4.73 cust/min)/(3x 1.77 cust/min) = 0.89

CV2 2 2(s+1)−1
a +CVp Uklizakon Wp
Wq = × × =1.40 mins
2 1−Uklizakon s

• Alternative 2: Three separate queues

Utilization = (4.73/3 cust/min)/(1.77 cust/min) = 0.89

CV2a +CV 2
p Uklizakon
Wq = × ×Wp =4.61 mins
2 1−Uklizakon

Conclusion: Alternative 1 is better

28
Example: Pooling

Alternative 1: Common Queue Alternative 2: Separate Queues

Some servers might be idle even if some

people are waiting in line!
29
Limitations to Pooling
• Pooling may require workers to have a broader set of
skills, which may require more training and higher wages
• Imagine a call center that takes orders for McDonalds and
Wendys … now the order takers need to be experts in two sets
of menus
• Pooling may disrupt the customer–server relationship
• Patients like to see the same physician
• Pooling may increase the time-in-queue for one
customer class at the expense of another
• Removing priority security screening for first-class passengers
may decrease the average time-in-queue for all passengers but
will likely increase it for first-class passengers.

30
Learning Objectives: Session 10

• Economic implications: generating a staffing plan

• Waiting time formula for non-exponential distributions

• Impact of pooling: economies of scale

• Psychology of waiting

31
Psychology of Waiting and Suggestions
• Unoccupied time feels longer
than occupied time
• Try to divert the customer’s
attention when waiting
• Entertain, enlighten, and
engage (Katz et al., 1991)

• Pre-process waiting feels longer

than in-process waiting
• Ensure a quick “a foot in the door”
feeling to discourage any balking
or reneging

32
Psychology of Waiting and Suggestions
• Anxiety makes the waiting seem longer
• Ensure the waiting customers that they
are not forgotten
• Realize that new or infrequent customers
feel they wait longer than frequent users

• Uncertain waiting feels longer than

known, finite waiting
• Inform the customers of what and
how much longer to expect
• Surprise the customer with shorter
waiting time

33
Psychology of Waiting and Suggestions
• Unexplained waiting feels longer
than explained waiting
• Provide real reasons for keeping
customers waiting

• Unfair waiting feels longer than equitable waiting

• Redesign the waiting system to ensure fairness
• Maintain FCFS, if possible

34
Psychology of Waiting and Suggestions
• Uncomfortable waiting feels
longer than comfortable waiting
• Design a better waiting environment

• Solo waiting feels longer than

group waiting
• Create a waiting environment
that encourages interaction

35
Takeaways
• Pooling helps reducing waiting time
• Waiting time formula is useful when we do not know the
underlying distributions for inter-arrival and service
times
• Watch video for wrap-up
• https://www.youtube.com/watch?v=ptFlL2UaKkA&feature=youtu.be

• Next Class: Midterm exam review

36