Some Queuing Models

Managerial problems characterised by congestion

Telecommunication Where should switching equipment be placed and what should its
networks capacity be in order to provide an adequate level of service?

Clinics What are effects of reservations and appointments on throughput of

patients?
Manufacturing cells Where are bottlenecks in production process and how can they be
alleviated?
ATM machines Which ones are overcrowded and by how much?
Parking administration How many permits should be given for a particular parking lot?

Playground How should waiting areas be designed to manage orderly flow of

customers at various attractions?
Retailing How many checkout points should there be in order to maintain a
high service level?
Manufacturing How large should a repair facility be to minimize time during which
production equipment is not available for use?

Mail order sales How many telephone lines/operators should be employed to hedge
against lost sales due to customer impatience?
 What is a "queuing system“?
• A "queuing system" consists of discrete objects, i.e., "items" that "arrive" at
some rate to the "system". Within the system the items may form one or
more queues and eventually receive "service" and exit.
Arrivals Departures
items in queue &
items in service

Flow of items through a queuing system

• While items are in system, they may be in queues, or may be in service or
some in queue and some in service. Interpretation depends on application
and goals of modeler.

Served Customers

Queueing System

Queue
C S
Customers CCCCCCC C S Service
C S facility
C S

Served Customers
 Queuing Model

Input: Waiting: Service: Output:

customers customers service customers
arrive wait for service is performed leave
to the systemin a single on one of the the system
FIFO queue s servers

Server 1

Queue Server 2
(Buffer) ●
●
●

Server s
System
 Typical components of queueing model

  Default

Size of calling population Infinite

inter-arrival time None (but we assume exponential)

Probability distribution of service time None (but we assume exponential)

Queue discipline FIFO (i.e., FCFS)

Size of waiting room Infinite

Number of parallel servers One server (but will consider several)

Options available to customer No balking.

Single / Multiple queues Single queue

Customer behaviour No reneging, no jockeying

 Queuing Model Parameters
• arrival rate (e.g. # of customers arriving per hour)
•• service
  rate of one server (e.g. # of customers per hour a server can serve)
• # of servers

• If arrival rate =, then average inter-arrival time =.

– Example: on average, 5 customers arrive per hour
customers/hour
min/customer – time between two consecutive arrivals

Watch out: in some situations “arrival rate ()” is given; in others, “average inter-arrival time ()” is
given.

• Similarly, if service rate = , then average service time = .

– Example: if one customer can be served in 10 minutes on average, then
0 min/customer,
customers/min customers/hour – we can serve 6 customers per hour

Watch out: in some situations “service rate ()” is given; in others, “average service time ()” is
given.
 Capacity, Demand, and Flow Rate
• If we have identical servers, each with service rate ,
•  
then the capacity of our system (total service rate) equals .
– Capacity =

• Arrival rate of customers, , is the demand for service.

– Arrival rate = Demand =

• Flow Rate is the actual number of customers served per unit of time.
– Flow Rate = min(Demand, Capacity) = min(, )

• If Demand Capacity (), capacity is sufficient to satisfy demand.

– In this case, Flow rate = Demand = .
– All customers can be served.

• If Demand Capacity (), capacity is insufficient for the demand.

– In this case, Flow rate = Capacity = .
– Some customers will not be served.

• All queuing models we will deal with assume: (i.e., sufficient capacity).
 A Deterministic Queue

•Graph
  of number of customers in system over time

No. of customers
2

1
1 2 3 4 5 6 7
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Time
Find:
• Avg. Flow Rate; Inter-arrival Time =
• Avg. Service Rate; Average Service Time =
• W = Avg. Time in System; = Avg. Time Waiting in Queue
• L= Avg. # of Units in System; = Avg. # of Units Waiting in Queue
• = Utilisation
• Capacity =
 A Deterministic Queue

•Graph
  of number of customers in system over time

No. of customers
2

1
1 2 3 4 5 6 7
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Time
Find:
• Avg. Flow Rate = 1/3; Inter-arrival Time = 3;
• Avg. Service Rate = 1/2; Average Service Time = 1/=2;
• W = Avg. Time in System = 2; = Avg. Time Waiting in Queue = 0;
• L= Avg. # of Units in System = 2/3 ; = Avg. # of Units Waiting in Queue = 0;
• = Utilisation = 2/3;
• Capacity = 1/2;
 Little’s Law: For any system/process operating in steady state
Arrival Departure
System/process
  𝐿= 𝜆 ×𝑊
 Average Number of Units in the System Average Flow Rate Average Time in System

 
[units of flow] [units of flow / units of time] [units of time]

•  
Examples
– On average, customers visit the bank every hour and spend 12 min 0.2 hours, on
average, in the bank. How many customers are there in the bank, on average?

Avg. # of customers in a bank

– On average 6,500 undergraduate students study at AUB. Average duration of studying is

3.5 years. How many students join AUB per year, on average?

Avg. Flow Rate .

 Little’s Law

•
• A Forqueuing system consists of a queue and a service.
system as a whole: …… (1)
Queue itself is a system; hence, = …… (2)
Customers will be either in queue (waiting) or in service: = + 1/ …… (3)
From (1) & (2),
) = = average number of busy servers = utilisation
• Remarkably, Little’s Law:
– Simple
– General: Surprisingly, no matter what:
• how many servers there are,
• whether each server has its own queue or a single queue feeds all servers,
• what service time distributions are,
• what distribution of inter-arrival times is,
• what the order of service of items is.
– Useful: If two of the terms are easy to estimate, then the Law quickly provides the
missing third value.
– Has numerous uses in Operations Management and Managerial Decision Making:
You have already encountered it:
Work-In-Process (WIP) = Throughput * Cycle Time
Queuing Model: One Server, Deterministic, Sufficient Capacity
•• In the simplest case, there is one server ()
 

Queue One Server

(Buffer)
System

• Example 1: Consider a system with server, customers per hour, customers

per hour, and no variability in arrival and service times:
Customer 5
Customer 4
Customer 3
Customer 2 Arrival
Waiting
Customer 1 Service
0 10 20 30 40 50 60

– There is no waiting!
 Queuing Model: One Server, Deterministic, Insufficient Capacity

• Example 2: Consider a system with server, customers per hour, customers per
• hour,
  and no variability in arrival and service times:
– Capacity is insufficient ()

Customer 5
Customer 4
Customer 3
Customer 2 Arrival
Waiting
Customer 1 Service
0 10 20 30 40 50 60

– Number of customers in the system is increasing without limit:

4
2
0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60

– Avg. waiting time , Avg. # of customers waiting

• With variability, number of customers would also increase without limit.
Queuing Model: One Server, Sufficient Capacity and Variability

• Example 3: Consider a system with server, customers per hour, customers per
• hour,
  and variable arrival time and/or service time
Customer 5
Customer 4
Customer 3
Customer 2 Arrival
Waiting
Customer 1 Service
0 10 20 30 40 50 60
– Number of customers in the system will go up and down:
4
2
0
0 4 8 12 16 20 24 28 32 36 40 44 48 52 56 60

• In most services, both arrivals and service times are random (variability exists).
• Due to variability, waiting will occur, even with sufficient capacity ().
 Exponential and Poisson Distributions

P r o b a b i l i ty D e n s i ty
•
• For many types of services, both inter-arrival times and service times are exponentially distributed.
 – Small times are
most-likely. Exponential Probability
Density Function
8.0

– The larger the time,

7.0
6.0
5.0
4.0

the smaller its likelihood. 3.0

2.0
1.0

– Exponential distribution 0.0

0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90
Inter-arrival or Service Time

is defined completely by
one rate parameter: for arrival rate and for service rate

P r o b a b i l i ty
(mean = standard deviation = for inter-arrival times and for service times).

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Poisson Probability
Mass Function
Number of Customers Arriving per Unit of Time

• When inter-arrival times are exponentially distributed with mean = , number of

customers arriving per unit of time has Poisson distribution with mean = variance = .
– We say that the arrival process is Poisson.
 Memoryless Property of Exponential Random
Variables
•  
• Property 1: An exponential random variable is
memoryless.

And the converse is true as well:

• Property 2: A memoryless random variable is

exponentially distributed.

Conclusion: if is a continuous random variable,

then ‘memoryless’ and ‘has an exponential
distribution’ are equivalent statements.
 Kendall Notation

•• Kendall notation describes queuing models using symbols:

 
inter-arrival time distribution / service time distribution / #
of servers

• Possible distributions:
– M – exponential distribution (memoryless)
– G – general distribution (with a given mean and standard
deviation)
– D – deterministic (constant)

• We consider only the following models:

– M/M/s – exponential inter-arrival and service times
– M/G/1 and M/D/1 – exponential inter-arrival and one server with
general or deterministic service time
 M/M/1

•
• Exponential
•  arrival rate inter-arrival and service times, server
Average inter-arrival time
• service rate Average service time
• Utilization =
• Prob. of 0 customers in system
Fundamental Result (can be proven; depends on the M/M
assumption):
Avg. # of customers in system= ; rest follows:

• Avg. time in system

• Avg. time in queue
• Avg. # of customers in queue ;L= +
• Prob. of waiting
• Prob. of n customers in system = rn (1 – r)
• Prob. that waiting time in system exceeds t is P(W > t) = e–m (1–r)t for t ≥ 0
• prob. that waiting time in queue exceeds t is P(Wq > t) = r e–m (1–r)t for t ≥ 0

(Last 2 results need advanced mathematics)

Formulas work only if !

 M/M/s
•
• Exponential
•   inter-arrival and service times,
arrival rate Average inter-arrival time
servers

• service rate Average service time

• number of servers Total service rate

Let =

𝑃0 =¿ ¿
 
• Utilization

• Prob. of 0 customers in system

• Avg. # of customers in queue

• Avg. # of customers in system
• Avg. time in queue
• Avg. time in system
• Prob. of waiting
• Prob. of n customers in system
• Formulas work only if
 M/G/1
•
• Exponential
•
 arrival rate inter-arrival time, non-exponential service time,
Average inter-arrival time
server

• service rate Average service time

• standard deviation of service time

• Utilization

• Prob. of 0 customers in system

• Avg. # of customers in queue

• Avg. # of customers in system

• Avg. time in queue

• Avg. time in system

• Prob. of waiting
• Formulas work only if
• These formulas can be used for M/D/1 by assuming .
 Typical Managerial Questions
Service Quality
•  
• What is the average customer waiting time?
– Calculate .
– Note: if minutes, some customers might be waiting more than 5 minutes.

It is more appropriate to ask what percentage of customers wait less than 5 minutes, but this is
more difficult to calculate.

Capacity
• What is the average number of customers in the system or in the queue?
– Calculate or .
– Note: if customers, it is possible that at some point the number of waiting customers
might exceed 5.
• What is the probability that the number of customers in the system exceeds
5?

• What is the probability that the number of customers in the queue of M/M/2
system exceeds 5?
– With 2 servers, 5 customers are waiting when 2 + 5 = 7 customers are in the system.
–
 Dupit Corp. Problem

• Tech reps repair machines at customers’ sites.

• Current policy: Each rep’s territory is assigned machines so that
s/he will be active (repairing machines or travelling to site) 75%
of time.
– A repair call averages 2 hours.
– Machines average 50 workdays between repairs. Hence, 150
machines per rep. (3 repair calls per day)
• Proposed New Service Standard: average waiting time before a
rep begins trip to customer site should not exceed 2 hours.
In
•
Queuing terms:
  Customers: Machines needing repair.
• Arrivals: Calls to tech rep requesting repairs.
• Queue: Machines waiting at sites for repair to begin.
• Server: Tech rep.
• Service time: Total time representative is tied up (traveling to site or repairing).
Thus, a machine leaves queue and enters service when rep begins
trip to site.

• Assuming M/M/1: = 3 =0.75 (from M/M/1)

• Proposed New Service Standard: day
 Dupit Corp. Problem Suggested Approaches

• Approach 1: Decrease % of time reps are

expected to be repairing machines.
• Approach 2: Provide new equipment to
reduce time required for repairs.
• Approach 3: Adopt larger several-rep
territories, rather than 1-person
territories.
• Approach 4: Give owners of printer-
copier priority for receiving repairs over
others (Will not be analysed).
 Approach 1: Decrease % of time reps are expected to be repairing
machines

Effectively, lower rep’s utilization factor sufficiently: i.e.,

Lower r = l / m, until Wq ≤ 1/4 day,

 
Recall, is given = 4

Modify l = (# of machines assigned to rep)/50,

until you get required Wq.

Use Queuing calculator (M/M/1) to find that l = 2

gives = 0.25. Hence, assign to each rep a
territory with 100 machines (rather 150).
 Approach 2: Provide new equipment to reduce time required for
repairs

•New
  equipment would have following
effect on service-time distribution:
– Decrease mean (from 1/4 day to 1/5 day.

– Decrease standard deviation () from 1/4 day to 1/10 day.

Memoryless service time assumption not

justified ( ). Hence use M/G/1 calculator,
with = 3= 0.1 to find = 0.118.
 Approach 3: Adopt several-rep territories, rather than 1-person territories

• Experiment (M/M/s) with territories

with 2 reps, then 3, then ….. Until you
get desired result.
 
A territory with 3 tech reps:
Number of machines = 450
l=9 m=4 s = 3 r = l/sm
= 0.75
to find = 0.1893
Comparison of Wq with Territories of Different
Sizes
Number Number
of of
Tech Machine
Reps s l m s r Wq
0.7
1 150 3 4 1 0.75 workday (6 hours)
5
0.7
2 300 6 4 2 0.321 workday (2.57 hours)
5
0.7
3 450 9 4 3 0.189 workday (1.51 hours)
5

1 0.7
4 600 4 4 0.127 workday (1.02 hours)
2 5

Conclusion: Diminishing returns –– First few consolidations are

most effective
 Some Managerial Insights Into Queueing
Systems
• A high utilization factor (workload) leads to surprisingly poor
performance.

See MMs.xlsx (2nd sheet)

• Decreasing variability of service times (without any change in mean)

improves performance substantially.

See MG1.xlsx (2nd sheet)

• Multiple-server queueing systems can perform satisfactorily with

somewhat higher utilization factors than can single-server queueing
systems. For example, pooling servers by combining separate single-
server queueing systems into one multiple-server queueing system
greatly improves measures of performance.

• Applying priorities when selecting customers to begin service can greatly

improve performance for high-priority customers.
 Economic Analysis of Number of Servers to Provide
  In many cases, consequences of making customers wait can be expressed as a
waiting cost.
 manager is interested in minimizing total cost.
TC = Expected total cost per unit time
SC = Expected service cost per unit time
WC = Expected waiting cost per unit time
objective is to choose number of servers so as to
Minimize TC = SC + WC
 When each server costs same (Cs = cost of server per unit time),
SC = Cs s
 When waiting cost is proportional to amount of waiting (Cw = waiting cost per unit
time for each customer), and assuming that the waiting cost is incurred while in
the queue and while in service:
WC = Cw L
However, if the waiting cost is incurred only while waiting to be served, then
WC = Cw
 Acme Machine Shop

• l = 120 customers/hour
m = 80 customers/hour
• Total cost of each server is $20/hour, i.e., Cs = $20.
• While servers busy, their value is $48/hour, i.e., Cw = $48.
• Choose s so as to Minimize TC = $20s + $48L.

See MMs_Economic_Analysis.xlsx