Waiting Times

Chapter 7

These slides are based in part on slides that come with Cachon & Terwiesch
book Matching Supply with Demand http://cachon-terwiesch.net/3e/. If you
want to use these in your course, you may have to adopt the book as a textbook
or obtain permission from the authors Cachon & Terwiesch.
1
utdallas.edu/~metin
 Learning Objectives
 Interarrival and Service Times and their variability
 Obtaining the average time spent in the queue
 Pooling of server capacities
 Priority rules

utdallas.edu/~metin
2
 Where are the queues?

utdallas.edu/~metin
3
 A Queue is made of a server and a queue in front

Processing
Buffer

Input: Queues Resources:

Passengers in an airport Check-in clerks at an airport
Customers at a bank Tellers at a bank
Patients at a hospital Nurses at a hospital
Callers at a call center Customer service representatives
(CRS) at a call center
Arrival rate
Capacity

We are interested in the waiting times in the queue and the queue length.

utdallas.edu/~metin
 An Example of a Simple Queuing System

Incoming Answered
calls Reps Calls
Calls processing
on Hold calls

 At peak, 80% of calls

Call center dialed received a busy
signal.
 Customers getting
through had to wait on
average 10 minutes
 Extra phone line
expense per day for
waiting was $25,000.
Blocked calls Abandoned calls
(busy signal) (tired of waiting) Financial consequences

Lost throughput Holding cost Cost of capacity $$$ Revenue $$$

Lost goodwill Cost per customer
Lost throughput (abandoned)
utdallas.edu/~metin
5
 A Somewhat Odd Service Process
Constant Arrival Rate (0.2/min) and Service Times (4 min)
Arrival rate 0.2/min = 1/(4 mins) = 1 every five minutes, which implies interarrival time of 5 minutes.
Units of arrival rate 1/min whereas units of interarrival time is min.
Arrival Service
Time Time
Patient

1 0 4
2 5 4
3 10 4
4 15 4
5 20 4
6 25 4
7 30 4
8 35 4
9 40 4
10 45 4
11 50 4
12 55 4

utdallas.edu/~metin 7:00 7:10 7:20 7:30 7:40 7:50 8:00 6

 Where is Variability?
 There certainly is significant (actually infinite) amount of waiting
when the arrival rate is greater than the service rate
– Equivalently, the processing capacity is less than the arrival rate
 More interestingly, variability can cause long waiting times.
Variability in
– Arrival process
– Processing times
– Availability of resources; Absent, sick, broken or vacationing servers.
– Types of customers; Priority versus regular customers.
– Routing of flow units; Recall the Resume Validation Example.
– Response of customers to waiting for a while; Wait more or abandon

utdallas.edu/~metin 7
 Variability: Where does it come from? Examples
Tasks:
• Inherent variation
• Lack of Standard Operating Procedures
• Quality (scrap / rework)

Buffer Processing
Input:
• Unpredicted Volume swings
• Random arrivals (randomness is the rule,
not the exception) Resources:
• Incoming quality • Breakdowns / Maintenance
• Product Mix • Operator absence
• Set-up times

Especially relevant in service operations

(what is different in service industries?):
Routes:
• emergency room • Variable routing
• air-line check in • Dedicated machines
• call center
• check-outs at cashier

utdallas.edu/~metin
 Random Arrival Rate and Service Times
Arrival Service Patient 1 Patient 3 Patient 5 Patient 7 Patient 9 Patient 11
Patient Time Time
Patient 2 Patient 4 Patient 6 Patient 8 Patient 10 Patient 12
1 0 5
2 7 6
3 9 7 Time

4 12 6 7:00 7:10 7:20 7:30 7:40 7:50 8:00

5 18 5
6 22 2
3
7 25 4
Number of cases

8 30 3
2
9 36 4
10 45 2
1
11 51 2
12 55 2
0
Averages 5 4 2 min. 3 min. 4 min. 5 min. 6 min. 7 min.
Interarrival
time Service times
utdallas.edu/~metin
9
 Variability Leads to Waiting Time
Average Arrival Rate (0.2/min) and Service Times (4 min)
Arrival Service
Patient Time Time
Service time
1 0 5
2 7 6
3 9 7
4 12 6
5 18 5
Wait time
6 22 2
7 25 4
8 30 3
9 36 4 5
10 45 2 4
11 51 2
3
12 55 2
2

1
Inventory
(Patients at lab) 0
7:00 7:10 7:20 7:30 7:40 7:50 8:00 10
utdallas.edu/~metin
 Is the incoming call rate stationary?
Number of customers
Per 15 minutes
160

140

120

100

80

60

40

20

0
Time
12:30
10:45

21:15

23:00
17:45

19:30
14:15

16:00
0:15

2:00

3:45

7:15

9:00
5:30

utdallas.edu/~metin
11
 How to test for stationary?
Cumulative Cumulative
Number of Number of
Customers Customers

700 70

600 60

500 50

Expected arrivals
400 if stationary 40

300
30

200
Actual, cumulative 20

arrivals
100
10

0
0

6:00:00 7:00:00 8:00:00 9:00:00 10:00:00 7:15:00 7:18:00 7:21:00 7:24:00 7:27:00 7:30:00

Time Time

Not stationary over a day, try over an hour or over 30 minutes.

utdallas.edu/~metin
12
 Exponential distribution for Interarrival times

Percentage of calls with {Interarrival time  t}

100
1
Number of calls with interarrival time t

90

Probability{Interarrival time  t}
80 0.8

70

60 0.6

50

40 0.4

30

20 0.2

10

0
t time
0

Duration t

𝑡
Prob 𝐼𝐴 ≤ 𝑡 = 1 − exp − , IA interarrival time
𝑎
E 𝐼𝐴 = 𝑎, expected time between two arrivals in a row
StDev 𝐼𝐴 = 𝑎, expected and StDev are the same for exponential distribution
utdallas.edu/~metin
13
 Comparing empirical and theoretical distributions

Distribution Function
1

Exponential distribution
0.8

0.6

0.4
Empirical distribution

0.2

-
0
Interarrival time
0:00:00 0:00:09 0:00:17 0:00:26 0:00:35 0:00:43 0:00:52 0:01:00 0:01:09
utdallas.edu/~metin
14
 Analyzing the Arrival Process
 CVa=StDev(IA)/E(IA)
Stationary
Arrivals?

YES NO

Exponentially distributed Break arrival process up

inter-arrival times? into smaller time
intervals
YES NO

• Compute a: average interarrival time • Compute a: average interarrival time

• CVa=1 • CVa= St.dev. of interarrival times / a
• All results of chapters 8 and 9 apply • Results of chapter 8 or 9 do not apply,
require simulation or more complicated models

utdallas.edu/~metin
15
 Analyzing the Service Times
Seasonality and Variability
Call duration 2.5
[minutes] Week-end averages

1.5

1 Week-day averages

0.5

Time of the
0 day
0:00 23:00

800
Frequency

600

CVp: Coefficient of variation 400

of service times
200
Std. Dev = 141.46
Mean = 127.2
Call durations
0 N = 2061.00
[seconds]

utdallas.edu/~metin A
16
 Computing the expected waiting time Tq

Inventory Inventory
waiting Iq in service Ip

Inflow Outflow

Entry to Begin Departure

system Service

Waiting Time Tq Service Time p

Flow Time T=Tq+p

utdallas.edu/~metin
17
 Utilization

FlowRate 1 / a p
Utilizatio n  u   
Capacity 1 / p a

Example: Average Activity time=p=90 seconds

Average Interarrival time=a=300 seconds

Utilization=90/300=0.3=30%

utdallas.edu/~metin
18
 The Waiting Time Formula
Waiting Time Formula for Exponential Arrivals

 utilizatio n 
 aCV 2
 CV 2

Time in queue  Activity Time  
p

 1  utilizatio n   2 

Variability factor
Utilization factor
Service time factor

EX: Average Activity time=p=90 seconds; Average Interarrival time=a=300 seconds; CVa=1 and CVp=1.333
 0.3   1  1.333 
2 2
Average time in queue  90     53.57 sec
 1  0.3  2 

Waiting Time Formula above is a restatement of Pollaczek-Khinchin (PK) Formula:

1 1 𝑆𝑒𝑐𝑜𝑛𝑑 𝑚𝑜𝑚𝑒𝑛𝑡 𝑜𝑓 𝑎𝑐𝑡𝑖𝑣𝑖𝑡𝑦 𝑡𝑖𝑚𝑒

PK Formula: 𝑇𝑞 =
𝑎 1−𝑝/𝑎 2
𝑝/𝑎 1 𝑝2 +Variance of activity time
=𝑝
1−𝑝/𝑎 𝑝2 2
𝑝/𝑎 1+𝐶𝑉𝑝2 𝑢 1+𝐶𝑉𝑝2
=𝑝 =𝑝 19
utdallas.edu/~metin 1−𝑝/𝑎 2 1−𝑢 2
 Bank Teller Example
 An average of 10 customers per hour come to a bank teller who serves each
customer in 4 minutes on average. Assume exponentially distributed
interarrival and service times.
(a) What is the teller’s utilization?
(b) What is the average time spent in the queue waiting?
(c) How many customers would be waiting for this teller or would be serviced by this
teller on average?
(d) On average, how many customers are served per hour?
Deterministic Counterpart
Answer: p=4 mins; a=6 mins=60/10
p 4
a) u    0.66 Arrive Arrive
a 6
4 minutes 2 mins 4 minutes 2 mins
 0.66   1  1 
2 2
b) Tq  4     8
 1  0.66   2  Depart Depart

1 1
c) I   * Tq  p     * 8  4  2
a 6
d) If teller is busy wp2/3, outputs15 per hour.
If teller is idle wp 1/3, outputs0 per hour.
utdallas.edu/~metin
Average output  (2/3) *15  (1 / 3) * 0  10 per hour  Average input !!! 20
The Flow Time Increase Exponentially in Utilization

Average  p  Time in queue

 utilizatio n   CVa  CV p 
flow 2 2

time T  p  Activity Time    

 1  utilizatio n   2 


Increasing
Variability

Theoretical Flow Time

utdallas.edu/~metin Utilization 100% 21

 Computing Tq with m Parallel Servers

Inventory in the system I=Iq+ Ip Inventory

in service Ip

Inventory
waiting Iq
Inflow Outflow

Entry to system Begin Service Departure

Waiting Time Tq Service Time p

Flow Time T=Tq+p

utdallas.edu/~metin
22
 Utilization with m servers

FlowRate 1/ a p
Utilizatio n  u   
Capacity m * (1 / p) a m

Example: Average Activity time=p=90 seconds

Average Interarrival time=a=11.39 secs over 8-8:15
m=10 servers

Utilization=90/(10x11.39)=0.79=79%

utdallas.edu/~metin
23
 Waiting Time Formula for Parallel Resources

Approximate Waiting Time Formula for Multiple (m) Servers


 Activity time   utilizatio n
2 ( m 1) 1 
 CV 2
 CV 2

Time in queue    a p

 m   1  utilizatio n   2 


Example: Average Activity time=p=90 seconds

Average Interarrival time=a=11.39 seconds
m=10 servers
CVa=1 and CVp=1.333

 90   0.79
2 (101) 1 
  12
 1.333 2

Time in queue      24.94 sec

 10   1  0.79    2 
T  Tq  p  24.94  90  114.94 sec  1.916 min
utdallas.edu/~metin
24
 Online Retailer Example
Customers send emails to a help desk of an online retailer every 2
minutes, on average, and the standard deviation of the inter-arrival time
is also 2 minutes. The online retailer has three employees answering
emails. It takes on average 4 minutes to write a response email. The
standard deviation of the service times is 2 minutes.

(a) Estimate the average customer wait before being served.

(b) How many emails would there be -- on average -- that have been
submitted to the online retailer, but not yet answered?

Answer: a=2 mins; CVa=1; m=3; p=4 mins; CVp=0.5

a) Find Tq. b) Find Iq=(1/a)Tq

utdallas.edu/~metin
25
 Service Levels in Waiting Systems
Fraction of 1 90% of calls had to
customers who wait 25 seconds or
have to wait x less
seconds or less 0.8 Waiting times for those customers
who do not get served immediately

0.6

0.4 Fraction of customers who get served

without waiting at all

0.2

0
0 50 100 150 200

Waiting time [seconds]

• Target Wait Time (TWT)

• Service Level = Probability{Waiting TimeTWT}; needs distribution of waiting time
• Example: Deutsche Bundesbahn Call Center
- now (2003): 30% of calls answered within 20 seconds
- target: 80% of calls answered within 20 seconds
utdallas.edu/~metin
26
 Bank of America’s Service Measures

utdallas.edu/~metin
27
 Waiting Lines: Points to Remember
• Variability is the norm, not the exception
- understand where it comes from and eliminate what you can
- accommodate the rest

• Variability leads to waiting times although utilization<100%

• Use the Waiting Time Formula to

- get a quantitative feeling of the system
- analyze specific recommendations / scenarios

• Adding capacity is expensive, although some safety capacity is necessary

• Next case:
- application to call center
- careful in interpreting March / April call volume

utdallas.edu/~metin
 Summary of the formulas
1. Collect the following data:
- number of servers, m
- activity time, p
- interarrival time, a
- coefficient of variation for interarrival (CVa ) and processing time (CVp )

p
2. Compute utilization: u=a  m

3. Compute expected waiting time

 Activity time   utilizatio n
2 ( m 1) 1   
CVa2  CV p2
Tq    
 m   1  utilizatio n   2 
  

4. Based on Tq , we can compute the remaining performance measures as

Flow time T=Tq+p
Inventory in service Ip=m*u
Inventory in the queue=Iq=Tq/a
Inventory in the system I=Ip+Iq

utdallas.edu/~metin
29
 Staffing levels
Cost of direct labor per serviced unit

Total wages per time m x (wages per time) p x (wages per time)
Cost of Direct Labor   
Flow rate  Arrival rate 1/a u
Because ma=p/u

Ex: $10/hour wage for each CSR; m=10

Activity time=p=90 secs; Interarrival time=11.39 secs
1-800 number line charge $0.05 per minute
p 90
Utilization  u    0.79
m x a 10x11.39
1.5 min/call x (16.66 cents/min)
Cost of Direct Labor   31.64 cents/call
0.79
Recall T  1.916; Cost of line charge per call  1.916x 0.05  $0.0958 / call

utdallas.edu/~metin
30
 Cost of direct labor per serviced unit
Another example with 9 servers

Ex: m=9. $10/hour wage for each CSR; Activity time=p=90 secs;
Interarrival time=11.39 secs; 1-800 number line charge $0.05 per minute

p 90
Utilization  u    0.878
m x a 9x11.39
1.5 min/call x (16.66 cents/min)
Cost of Direct Labor   28.474 cents/call
0.878

 90   0.878
2 ( 9 1) 1 

 12
 1.333 2

Time in queue      10 * 5.218 *1.39  72.54 sec
    
9 1 0.878  2 
T  Tq  p  72.54  90  162.54 sec  2.709 min

With T  2.709; Cost of line charge per call  2.709 * 0.05  $0.1354 / call

utdallas.edu/~metin 31
 Cost of Staffing Levels
m u Labor cost Line cost Total cost
per call charge per call per call
8 1.3458
9 0.878 0.2847 0.1354 0.4201
10 0.790 0.3164 0.0958 0.4122
11 0.4323
12 0.4593
13 0.4887
14 0.5193
15 0.5503
The optimal staffing level m=10
utdallas.edu/~metin 32
 Staffing Levels under various Interarrival Times

Number of customers Number of

CSRs
Per 15 minutes
160
17
16
140
15
14
120 13
12
100 11
10
80 9
8
60 7
6
40 5
4
20 3
2
0 1
Time
0:15

2:00

3:45

5:30

7:15

9:00

21:15

23:00
16:00

19:30
10:45

12:30

14:15

17:45

utdallas.edu/~metin
33
 The Power of Pooling
Independent Resources Waiting
Time Tq
2x(m=1)
70.00
m=1
60.00

50.00

40.00
m=2
30.00

20.00
m=5
Pooled Resources 10.00 m=10
(m=2)
0.00
60% 65% 70% 75% 80% 85% 90% 95%
Utilization u
Implications:
+ balanced utilization

+ Shorter waiting time (pooled safety capacity)

utdallas.edu/~metin - Change-overs / set-ups

 Service-Time-Dependent Priority Rules
• Flow units are sequenced in the waiting area (triage step)
• Shortest Processing Time (SPT) Rule
- Minimizes average waiting time
• Do you want to wait for a short process or a long one?

Service times:
A C A: 9 minutes
B: 10 minutes
B 4 min. D C: 4 minutes
9 min.
D: 8 minutes
19 min. C 12 min. A

23 min. D 21 min. B

Total wait time: 9+19+23=51min Total wait time: 4+12+21=37 min

• Problem of having “true” processing times

utdallas.edu/~metin
 Service-Time-Independent Priority Rules
•Sequence based on importance
- emergency cases; identifying profitable flow units
•First-Come-First-Serve
- easy to implement; perceived fairness
•The order in which customers are served does Not affect the average waiting time.
•W(t): Work in the system
•An arrival at t waits until the work W(t) is completed
W(t)

p2

p1

First Second Last Time t

utdallas.edu/~metin arrival arrival departure
 Service-Time-Independent Order does not
affect the waiting time
Order: 1-1-2 Order: 1-2-1

W(t) W(t)

First finishes
Time t Second finishes Time t

Work is conserved even when the processing order changes.

No matter what the order is, the third arrival finds the same amount of work W(t).

utdallas.edu/~metin
 Summary
 Interarrival and Service Times and their variability
 Obtaining the average time spent in the queue
 Pooling of server capacities
 Priority rules

utdallas.edu/~metin
38