Waiting Line Management

Ravindra S. Gokhale

Examples of Waiting Lines

Customers waiting at the single window counter of a bank Customers waiting at a restaurant for food to be served (after placing the order) Children waiting at an amusement park for a particular ride Documents waiting for their turn to be processed Data network where packets arrive, wait in various queues, receive service

at various points, and exit after some time (computer networking)

A telephone call waiting to get through in a busy telecommunication network Jobs waiting for their turn to be processed on a CNC machine

Machines waiting for repairs

Early Studies on Waiting Line Problems

A. K. Erlang, a Danish engineer who worked for the Copenhagen Telephone Exchange, published the first paper on queuing theory in 1909

David G. Kendall introduced an A/B/C queuing notation in 1953

It was later on extended as A/B/C/K/N/D notation

Leonard Kleinrock worked on queuing theory in the early 1960s

His work is used in modern packet switching networks

John Little published the proof of Littles Law in 1961

Structure of the Waiting Line Problem

Arrival process

It is the input process

The distribution that determines how the tasks arrives in the system
May be deterministic (example: prior appointment) or probabilistic Most common distribution for modeling arrivals: Poisson Arrivals are known as customers

Structure of the Waiting Line Problem (cont)

Service process

Describes the processing of tasks The distribution that determines how the tasks leave the system May be deterministic (example: computerized controlled process) or probabilistic Most common distribution for modeling services: Exponential, Erlang

Structure of the Waiting Line Problem (cont)

Number of servers

Total number of parallel servers/stations available to process the task Single server, multiple servers In the basic models, the servers are assumed to be of equal capability

Structure of the Waiting Line Problem (cont)

Capacity of the queue

Number of arrivals that the queue can hold Assumed to be infinite for the basic models

Structure of the Waiting Line Problem (cont)

Calling population

Population of potential arrivals Infinite (example: shoppers arriving at a shopping mall) or finite (example: machines arriving for maintenance) For finite (and small) population, the effective arrival rate decreases after each arrival and increases after each service completion

Structure of the Waiting Line Problem (cont)

(Queue) Discipline

FCFS First come first served SPT Shortest processing time first EDD Earliest due date first SIRO Service in random order Pre-emptive Example: In healthcare systems, a higher priority arrival can interrupt some established discipline or even some ongoing service, due to medical emergency

Nomenclature
Kendalls notation
A/B/C/K/N/D

A = Arrival distribution B = Service distribution C = Number of servers K = Capacity of the system (considered infinite if not specified) N = Calling population (considered infinite if not specified) D = Discipline of the queue (considered as FCFS if not specified)

Examples

M/M/1///FCFS (denoted just as: M/M/1) M/D/1///FCFS (denoted just as: M/D/1) M/M/s///FCFS (denoted just as: M/M/s) M/M/s//N/FCFS
10

Customer Behavior
Patient customer

Joins the queue and waits till his/her turn for servicing

Impatient customer

Balking Customer arrives, but decides not to join the queue due to longer queue length Reneging After being in the queue for some time, the customer leaves the system, without taking the service Jockeying A customer switches from one queue to another anticipating faster service

In this course we will assume only patient customer behavior

11

A queue forms not just because the arrival rate is greater than the service rate*, but because the nature of arrivals (and service times) is probabilistic.
*

The systems where arrival rate is greater than the service rate (and the
calling population is infinite) are unstable systems and cannot be sustained after some point of time.

12

Different Types of Systems

Based on channels and phases

Channel: Number of options available for the customer Phase: Number of stages in the service Number of phases Single
One channel sufficient and service offered in a single stage

Multiple
One channel sufficient, but service can best be offered in multiple stages

Number of channels

Single

Single server hair cutting shop Single server cobbler shop

Small or medium grocery shop with different stages placing order, getting goods, paying cash

Multiple

Multiple channels necessary and but service requires single stage

Multiple channels necessary and service requires different stages

Single window counter in a bank with multiple tellers Toll plaza

Automobile assembly line with multiple lines and multiple workstations Large petrol filling station
13

Performance Measures for Analysis (Quantitative)

Average number of customers in the queue (Lq)

Waiting to be served

Average number of customers in the system (Ls)

Includes those waiting to be served as well as those being served

Average waiting time in the queue (Wq)

Waiting time before commencement of service

Average waiting time in the system (Ws)

Includes the total time i.e. from joining the queue till completion of service

Service facility utilization ()

Signifies how busy is the service channel(s)

14

Littles Law
The long-term average number of customers in a stable system Ls is equal to the long-term average arrival rate, , multiplied by the long-term average

time a customer spends in the system, Ws

Ls = Ws Holds good for a wide variety of systems Does not require knowledge about distribution of arrivals and service times Littles Law cannot be applied to systems with finite calling population

15

Littles Law
Practical applications:

Manufacturing If one measures the average time a product spends in the system i.e. Ws and the rate of arrival of products i.e. , one can compute the Work in Process (WIP) Inventory i.e. Ls

Services If one counts the average number of customers in a system Ls and the rate of arrival of customers , then one can compute the average waiting time spent by a customer in the system Ws

Littles Law cannot be applied to systems with finite calling population

16

Single Server Model : M/M/1

Arrival rate follows a Poisson distribution

That means, time between the arrivals follows an Exponential distribution Arrival rate is denoted by customers per unit time

Service times follow an Exponential distribution

That means, service rate follows a Poisson distribution Service rate is denoted by customers per unit time

For the system to be stable, >

17

Single Server Model : M/M/1 (cont)

Server utilization: = / Probability of n customers in the system: Pn = (1 ) n

Probability that there are no customers in the system: P0 = (1 )

(This is, the probability that the server is idle)

Average number of customers in the queue: Lq = 2/[ ( )] Average waiting time in the queue: Wq = Lq /

Average number of customers in the system: Ls = Lq + ( / )

Average waiting time in the system: Ws = Ls / = Wq + (1 / )

18

Single Server Model : M/M/1 (cont)

Numerical example
Customer arrival at a single server post office counter is Poisson with an average rate of 6 customers per hour. The service times are exponential.

The server is idle for 20% of the time, on an average.

a. Determine the average service time, average length of the queue, average waiting time before being served, average time spent in the system and average number of people in the system. b. What is the probability that there are more than 4 customers in the system? c. What is the desired average service time if the management wants to ensure that only for 10% of the time, the number of customers in the system exceed four?

19

Single Server Model with Deterministic Service Times : M/D/1

Arrival rate follows a Poisson distribution

That means, time between the arrivals follows an Exponential distribution Arrival rate is denoted by customers per unit time

Service times are deterministic

Applicable in situations where the service process is automated Service rate is denoted by customers per unit time

For the system to be stable, >

20

Single Server Model with Deterministic Service Times : M/D/1 (cont)

Server utilization: = / Probability that there are no customers in the queue: P0 = (1 )

(This is, the probability that the server is idle)

Average number of customers in the queue: Lq = 2/[2 ( )] Average waiting time in the queue: Wq = Lq /

Average number of customers in the system: Ls = Lq + ( / ) Average waiting time in the system: Ws = Ls / = Wq + (1 / )

Numerical examples

21

Multiple Server Model : M/M/s

Arrival rate follows a Poisson distribution

That means, time between the arrivals follows an Exponential distribution Arrival rate is denoted by customers per unit time

Service times follow an Exponential distribution

That means, service rate follows a Poisson distribution Service rate is denoted by customers per unit time

Number of servers = s
For the system to be stable, (s ) >

22

Multiple Server Model : M/M/s (cont)

Server utilization: = / (s ) Probability that an arrival has to wait:
s

Pw

1 P0 s!

s s

(This is, the probability that all the servers are busy)
where P0 = Probability that there are no customers in the system, that is, probability that all servers are idle

Average number of customers in the system:

L s P0

1 [s -1]!

[s ]2

Average waiting time in the system: Ws = Ls / Average number of customers in the queue: Lq = Ls - ( / ) Average waiting time in the queue: Wq = Lq / = Ws - (1 / )
23

Multiple Server Model : M/M/s (cont)

Numerical example 1
Customer arrival at a bank is Poisson with an average rate of 24 customers per hour. There are three counters in the bank and each service counter has exponential service time with an average of 6 minutes per customer. With this data, there is a 6% probability that all the servers are idle. a. Determine the service facility utilization, average length of the queue, average waiting time before being served, average time spent in the system and average number of people in the system. b. What is the probability that all the servers are busy? c. If the efficiency of the service facility is improved and the average service time is brought down to 5 minutes per customer, can the

management have only two counters for the bank?

24

Multiple Server Model : M/M/s (cont)

Numerical example 2
Maintenance department in a large manufacturing organization has the responsibility to look after the maintenance of all the machines in the organization. Assume that all the machines work round the clock. The maintenance department functions as independent crews, which are assumed to be equally capable. Each crew has three personnel involved. Currently there are two such crews. The time between failure of machines follows an exponential distribution, and on an average 2 machines fail everyday. The repair time of a maintenance crew for one machine is exponentially distributed with a mean of 20 hours. With this information, there is a 0.09 probability that both the crews are idle. The average idle time cost of a machine is Rs. 300 per hour. Each maintenance personnel is paid Rs. 75 per hour. What is the average cost of each breakdown?

25

Single Server Model with Finite Calling Population: M/M/1//N/FCFS

Arrival rate follows a Poisson distribution

That means, time between the arrivals follows an Exponential distribution

Arrival rate is denoted by customers per unit time

Service times follow an Exponential distribution

That means, service rate follows a Poisson distribution Service rate is denoted by customers per unit time

Calling population is finite = N customers Littles Law cannot be applied in this case

26

Single Server Model with Finite Calling Population: M/M/1//N/FCFS (cont)

Average number of customers in the queue:

Lq N

(1 P0 )

where P0 = Probability that there are no customers in the system, that is, the server is idle

Average number of customers in the system: Ls = Lq + (1 P0)

Average waiting time in the queue: Wq = Lq /[ (N - Ls)] Average waiting time in the system: Ws = Wq + (1 / )

27

Single Server Model with Finite Calling Population: M/M/1//N/FCFS (cont)

Numerical example
XYZ Manufacturing Company has 20 machines. Each machine operates an

average of 200 hours before breaking down. The maintenance staff

includes 5 workers and works together as a team. The average time to repair is 3.6 hours. Breakdown rate is Poisson distributed and service time is exponentially distributed. With this data, the probability that the

maintenance staff is idle turn out to be 0.65. Find the average number of
machines in the repair queue, average number of machines in the repair system, average waiting time in the repair queue and average waiting time in the repair system.

28

Behavioral and Other Considerations

Non-linear waiting costs and the psychology of waiting Customers have some threshold for waiting Nature of emergency Social justice Perception of inequality Environment

How pleasant is the waiting environment

Service providers can modify the environment so that waiting may not seem long Feedback and accurate information about expected waiting time

29

Some Mitigation Techniques

Allow customers to serve themselves (at least partially) Shorten service times by performing tasks in advance Use two queues for each server Reward customers for being patient and waiting

30