Operations

Management
Waiting-Line Models
Module D

D-1
 Outline
Characteristics of a Waiting-Line System.
 Arrival characteristics.
 Waiting-Line characteristics.
 Service facility characteristics.

Waiting Line (Queuing) Models.

 M/M/1: One server.
 M/M/2: Two servers.
 M/M/S: S servers.
 Cost comparisons.

D-2
 Waiting Lines
 First studied by A. K. Erlang in 1913.
 Analyzed telephone facilities.
 Body of knowledge called queuing theory.
 Queue is another name for waiting line.
 Decision problem:
 Balance cost of providing good service with cost
of customers waiting.

D-3
 You’ve Been There Before!

Thank you for holding.

Hello...are you there?
The average person
spends 5 years waiting
in line!!

‘The other line

always moves faster.’
© 1995 Corel Corp.

D-4
 Waiting Line Examples
Situation Arrivals Servers Service Process

Bank Customers Teller Deposit etc.

Doctor’s Patient Doctor Treatment

office

Traffic Cars Traffic Controlled

intersection Signal passage

Assembly line Parts Workers Assembly

D-5
 Waiting Line Components
 Arrivals: Customers (people, machines, calls, etc.)
that demand service.
 Waiting Line (Queue): Arrivals waiting for a free
server.
 Servers: People or machines that provide service
to the arrivals.
 Service System: Includes waiting line and servers.

D-6
Car Wash Example

D-7
 Key Tradeoff

 Higher service level (more servers, faster

servers)

 Higher costs to provide service.

 Lower cost for customers waiting in line

(less waiting time).

D-8
 Waiting Line Terminology
 Queue: Waiting line.
 Arrival: 1 person, machine, part, etc. that
arrives and demands service.
 Queue discipline: Rules for determining the
order that arrivals receive service.
 Channels: Parallel servers.
 Phases: Sequential stages in service.

D-9
 Input Characteristics
 Input source (population) size.
 Infinite: Number in service does not affect
probability of a new arrival.
 A very large population can be treated as infinite.
 Finite: Number in service affects probability of a
new arrival.
 Example: Population = 10 aircraft that may need repair.

 Arrival pattern.
 Random: Use Poisson probability distribution.
 Non-random: Appointments.
D-10
 Poisson Distribution
 Number of events that occur in an interval of time.
 Example: Number of customers that arrive each half-hour.

 Discrete distribution with mean = 

 Example: Mean arrival rate = 5/hour .
 Probability: e  x
P( x ) 
x

 Time between arrivals has a negative exponential

distribution.

D-11
 Poisson Probability Distribution

0.30 0.30

0.25 0.25

0.20 0.20
Probability

0.15 0.15

Probability
0.10 0.10
0.05
0.05
0.00
0 1 2 3 4 5 6 7 8 9 10 11 12 0.00
x
0 1 2 3 4 5 6 7 8 9 10 11 12
x

=2 =4

D-12
 Behavior of Arrivals
 Patient.
 Arrivals will wait in line for service.

 Impatient.
 Balk: Arrival leaves before entering line.
 Arrival sees long line and decides to leave.
 Renege: Arrival leaves after waiting in line a while.

D-13
 Waiting Line Characteristics
 Line length:
 Limited: Maximum number waiting is limited.
 Example: Limited space for waiting.
 Unlimited: No limit on number waiting.
 Queue discipline:
 FIFO (FCFS): First in, First out. (First come, first served).
 Random: Select arrival to serve at random from those
waiting.
 Priority: Give some arrivals priority for service.

D-14
 Service Configuration
 Single channel, single phase.
 One server, one phase of service.
 Single channel, multi-phase.
 One server, multiple phases in service.
 Multi-channel, single phase.
 Multiple servers, one phase of service.
 Multi-channel, multi-phase.
 Multiple servers, multiple phases of service.
D-15
 Single Channel, Single Phase
Service system
Served
Arrivals Queue units
Service
facility

Ships at Ship unloading system Empty

sea Waiting ship line ships
Dock

D-16
 Single Channel, Multi-Phase
Service system
Served
Arrivals Queue units
Service Service
facility facility

Cars McDonald’s drive-through Cars

in area Waiting cars & food
Pay Pick-up

D-17
 Multi-Channel, Single Phase
Service system
Served
units
Service
Arrivals Queue facility

Service
facility

Example: Bank customers wait in single line for one

of several tellers.
D-18
 Multi-Channel, Multi-Phase
Service system
Served
units
Service Service
Arrivals Queue facility facility

Service Service
facility facility

Example: At a laundromat, customers use one of several

washers, then one of several dryers.
D-19
 Service Times
 Random: Use Negative exponential probability
distribution.
 Mean service rate = 
 6 customers/hr.
 Mean service time = 1/
 1/6 hour = 10 minutes.

 Non-random: May be constant.

 Example: Automated car wash.

D-20
 Negative Exponential Distribution
.4
 Continuous distribution. =1
 Probability: =2

Probability t>x
.3
f ( t  x )  e μ x =3
.2 =4
 Example: Time between
arrivals. .1
 Mean service rate = 
 6 customers/hr. 0.
 Mean service time = 1/ 0 2 4 6 8 10
 1/6 hour = 10 minutes x

D-21
 Assumptions in the Basic Model
 Customer population is homogeneous and infinite.
 Queue capacity is infinite.
 Customers are well behaved (no balking or reneging).
 Arrivals are served FCFS (FIFO).
 Poisson arrivals.
 The time between arrivals follows a negative exponential
distribution

 Exponential service times: Services are described by the

negative exponential distribution.
D-22
 Steady State Assumptions
 Mean arrival rate , mean service rate , and
the number of servers are constant.

 The service rate is greater than the arrival rate.

 These conditions have existed for a long time.

D-23
 Queuing Model Notation
a/b/S
Number of servers or channels.
Service time distribution.
Arrival time distribution.

 M = Negative exponential distribution (Poisson

arrivals).
 G = General distribution.
 D = Deterministic (scheduled).
D-24
 Types of Queuing Models
 Simple (M/M/1).
 Example: Information booth at mall.
 Multi-channel (M/M/S).
 Example: Airline ticket counter.
 Constant Service (M/D/1).
 Example: Automated car wash.
 Limited Population.
 Example: Department with only 7 drills that may
break down and require service.
D-25
 Common Questions
 Given ,  and S, how large is the queue (waiting
line)?
 Given  and , how many servers (channels) are
needed to keep the average wait within certain
limits?
 What is the total cost for servers and customer
waiting time?
 Given  and , how many servers (channels) are
needed to minimize the total cost?

D-26
 Performance Measures
 Average queue time = Wq
 Average queue length = Lq
 Average time in system = Ws
 Average number in system = Ls
 Probability of idle service facility = P0
 System utilization = 
 Probability of more than k units in system = Pn > k
 Also, fraction of time there are more than k units in the system.

D-27
General Queuing Equations

=
S

1
=
Ws Wq + 
Given one of Ws , Wq ,

Ls = Lq +  Ls, or Lq you can use
these equations to
Lq =  W q find all the others.

Ls =  W s

D-28
 M/M/1 Model
 Type: Single server, single phase system.
 Input source: Infinite; no balks, no reneging.
 Queue: Unlimited; single line; FIFO (FCFS).
 Arrival distribution: Poisson.
 Service distribution: Negative exponential.

D-29
 M/M/1 Model Equations

Average # of customers in system: L s =
 -
1
=
Average time in system: W s  - 

2
Average # of customers in queue: L q =
 ( -  )

Average time in queue: Wq = 

 ( -  )

System utilization =

D-30
 M/M/1 Probability Equations

Probability of 0 units in system, i.e., system idle:


P0 = 1 -  = 1 -


Probability of more than k units in system:

()
 k+1
P = 
n>k

This is also probability of k+1 or more units in system.

D-31
 M/M/1 Example 1
Average arrival rate is 10 per hour. Average service
time is 5 minutes.

 = 10/hr and  = 12/hr

(1/ = 5 minutes = 1/12 hour)

Q1: What is the average time between departures?

5 minutes? 6 minutes?

Q2: What is the average wait in the system?

1
Ws = = 0.5 hour or 30 minutes
12/hr-10/hr
D-32
 M/M/1 Example 1
 = 10/hr and  = 12/hr

Q3: What is the average wait in line?

10
Wq = = O.41667 hours = 25 minutes
12 (12-10)

1
Also note: Ws = Wq + 

1 1 1
so Wq = W - = - = O.41667 hours
s  2 12

D-33
 M/M/1 Example 1
 = 10/hr and  = 12/hr

Q4: What is the average number of customers in line

and in the system?
102
Lq = = 4.1667 customers
12 (12-10)
10
Ls = = 5 customers
12-10

Also note: L q =  Wq = 10  0.41667 = 4.1667

L s =  Ws = 10  0.5 = 5

D-34
 M/M/1 Example 1
 = 10/hr and  = 12/hr

Q5: What is the fraction of time the system is empty

(server is idle)?

P0 = 1 -  = 1 - = 1 - 10 = 16.67% of the time
 12
Q6: What is the fraction of time there are more than 5
customers in the system?
10 6 = 33.5% of the time
P =
n>5 ( )
12

D-35
 More than 5 in the system...
Note that “more than 5 in the system” is the
same as:
 “more than 4 in line”
 “5 or more in line”
 “6 or more in the system”.

All are Pn>5

D-36
 M/M/1 Example 1
 = 10/hr and  = 12/hr

Q7: How much time per day (8 hours) are there 5 or

more customers in line?

P = 0.335 so 33.5% of time there are 6 or more in line.

n>5
0.335 x 480 min./day = 160.8 min. = ~2 hr 40 min.

Q8: What fraction of time are there 3 or fewer customers

in line?
10 5 = 1 - 0.402 = 0.598 or 59.8%
1-P =1-
n>4 ( )
12
D-37
 M/M/1 Example 2
Five copy machines break down at UM St. Louis per
eight hour day on average. The average service time for
repair is one hour and 15 minutes.

 = 5/day ( = 0.625/hour)

1/ = 1.25 hours = 0.15625 days

 = 1 every 1.25 hours = 6.4/day

Q1: What is the average number of “customers” in the

system?
5/day
LS = = 3.57 broken copiers
6.4/day-5/day
D-38
 M/M/1 Example 2
 = 5/day (or  = 0.625/hour)
 = 6.4/day (or  = 0.8/hour)

Q2: How long is the average wait in line?

5
Wq = = 0.558 days (or 4.46 hours)
6.4(6.4 - 5)

0.625
Wq = = 4.46 hours
0.8(0.8 - 0.625)

D-39
 M/M/1 Example 2
 = 5/day (or  = 0.625/hour)
 = 6.4/day (or  = 0.8/hour)

Q3: How much time per day (on average) are there 2 or
more broken copiers waiting for the repair person?

2 or more “in line” = more than 2 in the system

5 3
P
n>2 = ( )
6.4
= 0.477 (47.7% of the time)

0.477x 480 min./day = 229 min. = 3 hr 49 min.

D-40
 M/M/1 Example 3
A coffee shop sees on average one arrival every two minutes
in the morning. The average service time (for preparing the
drink, paying, etc.) is 90 seconds. If you leave your house at
9:00 am, and it is a 10 minutes drive to the coffee shop, and
then a 15 minute drive to school, what time would you expect
to arrive at school (on average)?

 = 30/hr  = 40/hr = 1 every 90 seconds

30
Wq = = 0.075 hrs = 4.5 minutes
40(40 - 30)

Arrival at school on average is at 9:31 am

(9:00 am + 10 min + 4.5 min + 1.5 min + 15 min = 9:31 am)
D-41
 M/M/S Model
 Type: Multiple servers; single-phase.
 Input source: Infinite; no balks, no reneging.
 Queue: Unlimited; multiple lines; FIFO (FCFS).
 Arrival distribution: Poisson.
 Service distribution: Negative exponential.

D-42
 M/M/S Equations
1
Probability of zero people or P0 
M 1 1  λ  n  1  λ  M Mμ
units in the system:        
 n 0 n!  μ   M!  μ  Mμ  λ
 
M
  
Average number of people or   
Ls  P
units in the system: M  !M    

M
  
Average time a unit spends   
Ws  P 
in the system: M  !M    

Note: M = number of servers in these equations

D-43
 M/M/S Equations

Average number of people or 

units waiting for service: L q  Ls 


Average time a person or 

Wq  Ws 
unit spends in the queue: 

D-44
 M/M/2 Model Equations

Average time in system: 4

Ws =
42 - 2

Average time in queue: Wq =

 2

 (2 + )(2 -)

Average # of customers in queue: Lq =  Wq

Average # of customers in system: Ls =  Ws

Probability the system is empty: 2 -

P0 =
2 + 

D-45
 M/M/2 Example
Average arrival rate is 10 per hour.
Average service time is 5 minutes for each of 2 servers.

 = 10/hr,  = 12/hr, and S=2

Q1: What is the average wait in the system?

412
Ws = = 0.1008 hours = 6.05 minutes
4(12) -(10)
2 2

D-46
 M/M/2 Example
 = 10/hr,  = 12/hr, and S=2

Q2: What is the average wait in line?

(10)2
Wq = = 0.0175 hrs = 1.05 minutes
12 (212 + 10)(212 - 10)

1
Also note: Ws = Wq + 

1
so Wq = W -
s  = 0.1008 hrs - 0.0833 hrs = 0.0175 hrs

D-47
 M/M/2 Example
 = 10/hr,  = 12/hr, and S=2

Q3: What is the average number of customers in line

and in the system?

L q =  Wq = 10/hr  0.0175 hr = 0.175 customers

L s =  Ws = 10/hr  0.1008 hr = 1.008 customers

D-48
 M/M/2 Example
 = 10/hr and  = 12/hr

Q4: What is the fraction of time the system is empty

(server are idle)?
212 - 10
P0 = = 41.2% of the time
212 + 10

D-49
 M/M/1, M/M/2 and M/M/3
1 server 2 servers 3 servers
Wq 25 min. 1.05 min. 0.1333 min. (8 sec.)
= 0.417 hr = 0.0175 hr = 0.00222 hr
WS 30 min. 6.05 min. 5.1333 min.
Lq 4.167 cust. 0.175 cust. 0.0222 cust.
LS 5 cust. 1.01 cust. 0.855 cust.
P0 16.7% 41.2% 43.2%

D-50
 Waiting Line Costs
Service cost per day
= (# of servers) x (cost per day of each server) +
(# customers per day) x (marginal cost per customer)

Customer waiting cost per day

= (# of customers per day) x (average wait per customer)
x (time value for customer)

Time units must agree

D-51
 Service Cost per Day
 = 10/hr and  = 12/hr
Suppose servers are paid $7/hr and work 8 hours/day. Also,
suppose the marginal cost to serve each customer is $0.50.

M/M/1 Service cost per day

= $7/hr x 8 hr/day + $0.5/cust x 10 cust/hr x 8 hr/day
= $56 + $40 = $96/day

M/M/2 Service cost per day
= 2 x $7/hr x 8 hr/day + $0.5/cust x 10 cust/hr x 8 hr/day
= $112 + $40 = $152/day

D-52
 Customer Waiting Cost per Day
 = 10/hr and  = 12/hr
Suppose customer waiting cost is $10/hr.

M/M/1 Waiting cost per day

= 0.417 hr/cust x 10 cust/hr x 8 hr/day x $10/hr = $333.33/day

M/M/1 total cost per day= $96 + $333.33 = $429.33/day

M/M/2 Waiting cost per day

= 0.0175 hr/cust x 10 cust/hr x 8 hr/day x $10/hr = $14/day

M/M/2 total cost per day= $152 + $14 = $166/day

D-53
 Unknown Waiting Cost
Suppose customer waiting cost is not known = C.

M/M/1 Waiting cost per day

= 0.417 hr/cust x 10 cust/hr x 8 hr/day x C = 33.33C $/day
M/M/1 total cost per day = 96 + 33.33C

M/M/2 Waiting cost per day

= 0.0175 hr/cust x 10 cust/hr x 8 hr/day x C =1.4C $/day
M/M/2 total cost per day= 152 + 1.4C

M/M/2 is preferred when 152 + 1.4C < 96 + 33.33C or

C > $1.754/hr
D-54
 M/M/2 and M/M/3
Q: How large must customer waiting cost be for M/M/3 to
be preferred over M/M/2?

M/M/2 total cost = 152 + 1.4C

M/M/3 Waiting cost per day

= Cx 0.00222 hr/cust x 10 cust/hr x 8 hr/day = 0.1776C $/day
M/M/3 total cost = 208 + 0.1776C

M/M/3 is preferred over M/M/2 when

208 + 0.1776C < 152 + 1.4C
C > $45.81/hr
D-55
 Remember:  &  Are Rates

  = Mean number of If average service time is 15

minutes, then μ is 4
arrivals per time period. customers/hour
 Example: 3 units/hour.

  = Mean number of
arrivals served per time
period.
 Example: 4 units/hour.
 1/ = 15 minutes/unit.
D-56
 Other Queuing Models
 M/D/S
 Constant service time; Every service time is the same.
 Random (Poisson) arrivals.

 Limited population.
 Probability of arrival depends on number in service.
 Limited queue length.
 Limited space for waiting.

 Many others...
D-57
 Other Considerations
 Wait time & queue length increase rapidly for  >0.7
 Queue is small until system is about 70% busy; then
queue grows very quickly.
 Pooling servers is usually advantageous.
 Airport check-in vs. Grocery stores.
 Variance in arrivals & service times causes long waits.
 Long service times cause big waits.
 Cost of waiting is nonlinear.
 Twice as long wait may be more than twice as bad.
D-58
 More Considerations
 Reduce effect of waiting.
 Distract customers with something to do, look at
or listen to.
 Music, art, mirrors, etc.
 Provide feedback on expected length of wait.
 “Your call will be answered in 6 minutes…”

 Use self service can reduce load on servers.

 Self-service at grocery stores.

D-59