ISOM 2700: Operations Management

Session 9: Service Management I

Dongwook Shin
Dept. ISOM, HKUST Business School
Course Roadmap
Bottleneck
Little’s law
Utilization

Control chart
Acceptance sampling
Six sigma

Maximize
Profits

Decision tree method

Linear programming

Waiting time
Server utilization
1
What is Special About Service
Operations?

Capacity of service cannot be stored

and carried over!

2
The Process View (Previous Class)

Input Output

Waiting Work-in-process

Flow Time

Airline Waiting for Office Hours 3

Where the Time Goes?

In a life time, the average

American will spend--

SIX MONTHS: Waiting at stoplights

EIGHT MONTHS: Opening junk mail

ONE YEAR: Looking for misplaced objects

TWO YEARS: Unsuccessfully returning phone calls

FOUR YEARS: Doing housework

FIVE YEARS: Waiting in line

SIX YEARS: Eating

Source: US News & World Report, January 30, 1989, p. 81.

4
Managerial Implications of Long Waiting

5
Waiting Lines and Queueing Systems
• Where?
• Call centers
• Theme parks
• Fast food restaurant
• Supermarket
• …

• Questions:
• What are the root causes of waiting?
• How to reduce the waiting time?

6
Learning Objectives: Session 9

• Root causes of waiting in service operations

• Fundamentals of queueing models

7
Subway Sandwich at HKUST
• A server makes a sandwich exactly for 3 minutes
• Scenario 1: customers arrive exactly every 3 minutes

Buffer Server

• Scenario 2: customers arrive randomly (but with the

same average)

Buffer Server

Simulation Visualization
8
 Variability: Where Does It Come From?
Processing times:
• Inherent variation
• Lack of SOPs
• Quality (scrap / rework)

Task /
Buffer
Activity
Input:
• Random arrival
(randomness is
Resources:
the rule, not the
• Breakdown / Maintenance
exception)
• Operator absence
• Set-up times

Variability of arrival process Variability of service time

9
An Example of a Simple Queueing System

Call center

Incoming Answered
calls calls
Calls Sales staffs
on
Hold

Blocked calls: Abandoned calls:

At peak, 80% of Customers tired of Financial consequence:
calls receives a waiting abandons Loss in Revenue $$$
busy signal the queue

10
Root Causes of Waiting
• Why do queues exist?
• Variability of arrival process
• Variability of service time

• How to reduce waiting time?

• Reduce variability of arrival process
• Reduce variability of service time

• Examples: Disney land, Hospital

11
Learning Objectives: Session 9

• Root causes of waiting in service operations

• Fundamentals of queueing models

12
 Elements of Queueing Models

Service provided here

Customers leave the
Customers Arrive system
Servers
Queue

• Inter-arrival time: time between two consecutive arrivals (random)

…
Time
a1 a2 a3 a4 a5 a6 a7 …

• Service time: time taken to serve one customer

(random)
Time
s1 s2 s3 s4 s5 s6 s7
13
Modeling Random Inter-arrival and
Service Times: Exponential Distribution
• Inter-arrival time a ∼ ExponenMal(λ)
• Service time s ∼ ExponenMal(μ)
We call that
1
a random variable X
is exponentially
0.8
distributed with
rate 𝜆 if
0.6
𝑃 𝑋≤𝑡
P(X £ t)

= 1 − exp(−𝜆𝑡)
0.4

0.2 If X ~ Exponential(𝜆),
then E(X)=1/𝜆 and
0 Var(X)=1/𝜆!
Time t

14
Poisson Arrival Process

Inter-arrival time # of arrivals per period

Exponential
distribution
= Poisson
Distribution

Inter-arrival time Number of arrivals

~ Exponential(3) per unit time
~ Poisson(3)

15
 M/M/s Model
Service rate at each server
µ customers/min
Arrival rate
l customers/min Customers Experience /
System Performance
FCFS Servers
Queue

• Input of the queue

• Exponential inter-arrival time: arrival rate = l
• Exponential service time: service rate at each server = µ M/M/s
• Number of servers = s
queue
• Assumption: one queue; FCFS; l < sµ
• Output of the queue (System performance measures)
• Average number of customers in the queue Lq (or in the system L)
• Average time spent in the queue Wq (or in the system W)
• Probability of waiting
• Utilization of the server ρ
16
 Elements of Queueing Models

Service provided here

Customers leave the
Customers Arrive system
Servers
Queue

• Kendall’s notation

X/X/X

Number of servers
Service time distribution
Inter-arrival distribution
17
Remember: 𝜆 And 𝜇 Are Rates

• l = Average number of If average service time is

arrivals per time period 15 minutes, then μ is 4
customers/hour
• e.g., 3 units/hour

• µ = Average number of
people or items served per
time period
• e.g., 4 units/hour
• 1/ µ = 15 minutes/unit

18
 Server Utilization 𝜌
Service Rate 𝜇
per each server

Inter-arrival Rate 𝜆
Buffer s Servers

• Condition: 𝜆 < 𝑠𝜇
Flow Rate λ
UMlizaMon ρ= = < 100%
Capacity 𝑠µ
• (EX) There are 2 servers. The average arrival rate is 10
customers per hour. The average service time per
customer is 10 minutes. Server utilization = 83.33%

Server utilization is the average fraction of time a server is busy

19
 Stable and Unstable Queues

200 200
𝜆 = 1/35 𝜆 = 1/35
180 180
𝜇 = 1/120 𝜇 = 1/120
160 160

Total number of customers in the system

Total number of customers in the system

140
s=4 140
s=3

120
Utilization = 86% 120
Utilization = 114%
100 100

80 80

60 60

40 40

20 20

0 0
0 5000 10000 15000 20000 25000 30000 0 5000 10000 15000 20000 25000 30000
Time (seconds) Time (seconds)

20
Little’s Law Revisited
Waiting Time = 𝑇! Service Time = 𝑝

Inter-arrival Rate 𝜆 Transformation

Buffer Process Server

• Little’s law:
Avg. Number Flow Rate Flow Time
= ×
in System of System of System

21
Little’s Law Revisited

Inter-arrival Rate 𝜆
Single
Buffer
Server

Avg. Waiting Time Avg. Service Time

𝑊! 𝑊" = 1/𝜇

Avg. Number Waiting Avg. Number in Service

𝐿! = 𝜆𝑊! 𝐿" = 𝜆𝑊" = 𝜆/𝜇 = 𝜌

22
Formulas for M/M/1 Queueing Model
• Given: arrival rate = 𝜆, service rate = 𝜇
Condition
• Probability of exactly n customers in the system 𝜆<𝜇
𝑃" = 𝜌" (1 − 𝜌)
• Average number of customers in the system
𝜆 𝜌
𝐿= =
𝜇−𝜆 1−𝜌
• Average number of customers in queue
𝜆! 𝜌!
𝐿# = =
𝜇(𝜇 − 𝜆) 1 − 𝜌
• Average time in the system
𝐿 1
𝑊= =
𝜆 𝜇−𝜆
• Average time in the queue
𝐿# 𝜌
𝑊# = =
𝜆 𝜇−𝜆
23
Example: Pizza Hut
At a take-away only Pizza Hut
branch,

• Customers arrive randomly at the

average rate of 25 per hour

• There is one employee who can, on

average, serve one customer every 2
minutes

M/M/1 Queue

1 Employee
Exponential Service Time
Exponential Inter-arrival Time
24
Example: Pizza Hut (continued)
1) Average utilization of the employee
𝜆 25 𝑐𝑢𝑠𝑡/ℎ𝑟
= = 0.8333
𝜇 30 𝑐𝑢𝑠𝑡/ℎ𝑟

2) Average Queue Length

𝜆* 25*
𝐿) = = = 4.167
𝜇(𝜇 − 𝜆) 30(30 − 25)

3) Average number of customers in the system

𝜆 25
𝐿= = =5
𝜇 − 𝜆 30 − 25
25
Example: Pizza Hut (continued)

4) Average Waiting Time in Line

𝐿)
𝑊) = = 0.1667 ℎ𝑟𝑠 = 10 𝑚𝑖𝑛𝑠
𝜆

5) Average Amount of Time in the System

𝐿
𝑊 = = 0.2 ℎ𝑟𝑠 = 12 𝑚𝑖𝑛𝑠
𝜆

26
Congestion vs. System Utilization
100 For the M/M/1 model,
avg. number in the system
is:
𝜌
𝐿=
1−𝜌
10
𝜌 𝐿
8
𝐿 0 0
6
0.2 0.25
4
0.5 1
2
0.8 4
0 0.9 9
0 1.0
0.99 99
𝜌
0.999 999
27
Example: Pizza Hut Extended
At a take-away only Pizza Hut
branch,

• Customers arrive randomly at the

average rate of 25 per hour

• There is one employee who can, on

average, serve one customer every 2
minutes

If an extra employee (with the same training and hence the

same service rate) works in this branch, what will happen?

28
Example: Pizza Hut Extended
• Customers arrive randomly at the average rate of 25 per
hour
• There are 2 employees who can, on average, serve one
customer every two minutes

M/M/2 Queue

2 Employees
Exponential Service Time
Exponential Inter-arrival Time

29
Formulas for M/M/s Queueing Model
• Given: Arrival rate = 𝜆, Service rate = 𝜇 Condition:
𝜆 < 𝑠𝜇

• Probability of waiting
𝑠𝜇
𝑃$ = 𝐿# −1
𝜆
• Average number of customers in the system
𝜆
𝐿 = 𝐿# +
𝜇
• Average time in the system
𝐿
𝑊=
𝜆
• Average time in the queue
𝐿#
𝑊# =
𝜆
30
 Table for
Values of 𝐿)

31
Example: Pizza Hut Extended (continued)

• Number of customers in the queue

𝐿) ≈ 0.176

• Number of customers in the system

𝐿 ≈ 1.009

• Average waiting time in queue

𝐿) 0.176 𝑐𝑢𝑠𝑡
𝑊) = = = 0.007 ℎ𝑜𝑢𝑟 = 0.4 𝑚𝑖𝑛
𝜆 25 𝑐𝑢𝑠𝑡/ℎ𝑜𝑢𝑟

32
 Example: Pizza Hut Extended (continued)
• What is the benefit of adding an extra employee?

1 Employee 2 Employees
Probability of waiting in 0.833 0.245
line
Average Queue 4.167 0.176
Length
Average Number of
Customers in the system 5 1.009

Average Waiting 10 min 0.4 min

Time
33
Queueing System Cost Tradeoff
Cw = Cost of waiting per customer per hour
Lq = Mean number of waiting customers
Cs = Cost of service per server per hour
s = Number of servers

Total cost per hour = service cost + customer waiting cost

= 𝐶+ ×𝑠 + 𝐶, ×𝐿)

34
Queueing System Cost Tradeoff

35
Takeaways
• Variability causes congestion
• Reducing variability reduces waiting time
• M/M/s model to analyze average waiting time and the
number of customers in the system

Model parameters Performance measures

l
µ 𝜌, 𝑊# , 𝑊, 𝐿# , 𝐿

• Next class: More on queueing models

36