This paper appears in the Proceedings of the Sixth Youth Science Conference, 2000, Ministry of Education, Singapore.

DISCRETE–EVENT SIMULATION OF QUEUING SYSTEMS

Zhang Laifu Joel1 , Ng Wen Wei Jonathan Louis 1 and Tay Seng Chuan2
1
Anglo-Chinese School (Independent), 121 Dover Road, Singapore 119643.
2
Centre for Information Technology and Applications, Science Faculty, National University of Singapore,
3 Science Drive 2, Singapore 117543.

ABSTRACT
The type of queuing system a business uses is an important factor in determining
how efficient the business is run. In this project, we examine two types of queuing
systems: the single-channel and the multiple-channels queues which are commonly seen
in banks and fast food restaurants respectively. We use computer programs to simulate
the queues and predict the queue length, waiting time and wait probability. The input to
the simulation program is based on the statistics collected over a span of a week. The
discrete–event simulation approach is used to model the queuing systems and to analyze
the side effects when one system is changed to the other.

INTRODUCTION
As the size of the world’s population increases so do the number of queues and
their queue length. In the business world, more customers means more business
transactions. Out of the many ways to attract customers, an efficient queuing system
plays an important role as it reduces a customer’s waiting time. As a result, the shorter
waiting time makes customers happy, and one thing for sure is that a happy customer will
come back for business again.
This paper begins with a description of the single-channel and multiple-channels
queuing systems and their influence on a customer’s waiting time and wait probability1 .
We use discrete-event simulation program to verify the live data, and predict the
performance if the configuration of the existing queue is changed.

General Queue
In a queuing system, the calling population is assumed to be infinite [1]. That is, if
a unit leaves the calling population and joins the waiting line or enters service, there is no
change in the arrival rate 2 . Figure 1 shows the model used to analyze a general queue.
The arrivals occur one at a time in a random order and once the customer joins the
queuing system he will eventually receive the service.

Server

Waiting Line

Figure 1: A Simple Queue

1
The probability that a customer will have to join in a queue when he/she comes to a service station.
2
The effect of this assumption is negligible if the size of calling population is large.
This paper appears in the Proceedings of the Sixth Youth Science Conference, 2000, Ministry of Education, Singapore.

The arrival rate and services are modeled as variables which follow statistical
distributions. If the arrival rate is greater than the service rate, the waiting line will grow
without bound.

Single-Channel Queuing System

The single-channel queuing system can be seen in places such as banks and post
offices, where one single queue will diverge into a few counters. The moment a customer
leaves a service station, the customer at the head of the queue will go to the server. The
disadvantage of a single-channel queue is that the queue length seems to be very long,
thus it can discourage customers from joining the queue.

Waiting Line

Servers

Figure 2: Single-Channel Queuing System

Multiple-Channels Queuing System

The multiple-channels queuing system is commonly observed in fast food
restaurants like KFC, Burger King, McDonalds, etc. It is a system whereby the customers
line up in rows directly in front of each server. They are arranged in relatively straight
lines that do not converge. Generally, the customers in a multiple-channels queue feel
happier because the queue length is shortened as they are distributed to different count ers.

Waiting Lines Servers

Figure 3: Multiple-Channels Queuing System

Example of a Single-Channel Queue

POSB is a banking service entity that uses the single queue system. Customers
join one queue and when they reach the front they will go to a server that is idle. The
main characteristic of a single-channel queue is the first-come- first-serve feature.
This paper appears in the Proceedings of the Sixth Youth Science Conference, 2000, Ministry of Education, Singapore.

Example of a Multiple-Channels Queue

McDonalds is a chain of fast food restaurants that uses the multiple-channels
queue. The restaurants are centrally located in areas where many people pass through
daily like in Town Centres and around Orchard Road. In this type of queuing system the
customers can feel a sense of “speed” when they see less people standing in front of them
as compared to that in POSB.

MATERIALS AND METHODS

This project evaluates the performance of single-channel and multiple-channels
queuing systems, specifically that of fast food restaurants like McDonalds and banks like
POSB. To do this, we first went to both McDonalds and POSB to collect the statistical
data. We spent a week recording inter-arrival times and service times at different
locations and times.
Using the data collected as the input to the discrete-event simulation program, we
set up two simulators for predicting the behaviour of a single-channel queue and a
multiple-channels queue respectively.
The purpose of using simulation technique to analyze the collected data is to
avoid costly design errors, and to analyze the behaviours of the existing systems. More
importantly, simulation can be used to predict the performance of the existing system
when the input parameters such as the arrival rate and service rate are changed.
Simulation technique can also be applied to analyze the behaviours of system which has
not even been created yet [2].

Materials Used
The first phase of this project is data collection. We used stopwatches to time the
inter-arrival time and service time, and calculate the average timings as shown in Tables
1 and 2.

Table 1: Data collected at Choa Chu Kang and Clementi

MacDonalds from 15:30 to 16:30 and 10:30 to 11:30
Inter-Arrival Time (sec) Service Time (sec)
Average 48 70

Table 2: Data collected at Taman Warna and Clementi POSB

from 15:30 to 16:30 and 15:10 to 16:10 respectively
Inter-Arrival Time/s Service Time/s
Average 30 130

In the second phase, we used computer to analyze the collected data. A simulator
programmed in C++ was used for this purpose.
This paper appears in the Proceedings of the Sixth Youth Science Conference, 2000, Ministry of Education, Singapore.

Methods

Discrete–Event Approach
Discrete event is a technique used to model the real-world scenarios. In the
queuing model two types of events are used, namely arrival and departure [2]. The arrival
corresponds to the real- world event when a customer reaches a service station, and the
departure corresponds to the event when the customer leaves. Due to the causality
constraints, the arrival event for a customer must be executed before its departure event.
Each event has a timestamp corresponding to the wall-clock time when it occurs.
Discrete-event technique has been widely used in the simulation of communication and
transportation systems, such as telephone networks, seaport and airport operations, etc.

The Question: What if POSB changes its queue to multiple-channels and McDonalds
changes its queue to single-channel?
We first validate the data collected and compare the deviation of the observed
and predicated queue length, waiting time and wait probability. As the simulation results
follow closely the observed values, we brought up the question of whether McDonalds
and POSB will benefit if they switched their queuing systems. The simulator was used to
find out the answers.
The changes to the input data are as follows. To convert from a single-channel
queue to a multiple-channels queue for n servers we will have to divide the arrival rate by
n because the arriving customers are distributed to the n channels. Therefore, the arrival
rate to each channel is scaled down by a factor of n. As for converting a multiple-
channels queue to a single-channel que ue, the arrival rate is multiplied by n due to the
merging effect. The service rate remains unchanged for the conversions.

RESULTS
Tables 3 and 4 show the simulation results. As observed in both tables, the server
utilization remains fairly constant as the total workload does not changed. As for
converting from a multiple-channels queue to a single-channel queue, Table 3 shows that
the queue length is reduced by 48%. More significantly, the wait probability is reduced
from 0.68 to 0.35, and the average que ue time is reduced from 106 seconds to 11
seconds. This implies that the probability that a customer coming to the McDonalds will
have to wait for service is significantly reduced by half, and even if the customer is not
immediately served the waiting time is reduced by 9 folds. As such, the change from a
multiple-channels queue to single-channel will make a substantial improvement in the
queuing performance.

Table 3: Comparison of Queuing Performance for McDonalds

Queue Average Queue Ave. Queue Wait Server
Type Length Time (sec) Probability Utilization
Multiple-
Channels 1.52 106.36 0.68 0.69
(existing)
Single-
0.78 10.86 0.35 0.68
Channel
This paper appears in the Proceedings of the Sixth Youth Science Conference, 2000, Ministry of Education, Singapore.

On the other hand, Table 4 shows that the average queue length, average queue
time and the wait probability all become worse if the POSB is to convert its single-
channel queue to multiple-channels.
Table 4: Comparison of Queuing Performance for POSB
Queue Average Queue Ave. Queue Wait Server
Type Length Time (sec) Probability Utilization
Single-
Channel 4.25 127.34 0.68 0.86
(existing)
Multiple-
5.91 883.63 0.87 0.84
Channels

CONCLUSION
This paper has evaluated the performance of single-channel and multiple-channels
queues using the discrete-event simulation technique. The input to the simulators is based
on live data collected at McDonalds Fast Food Restaurant and POSB. Our simulation
results show that a single-channel queue is more efficient than a multiple-channels queue.
If so, why does McDonalds still want to use the multiple-channels queuing system? From
our observations, one reason is that in a multiple-channels queue the customers can hop
from one queue to another. Another reason is the impression of shorter waiting time but
our simulation result has proven it wrong. While hopping queues may seem to allow
customers to get to the server faster, it is not always true. A customer can hop to a shorter
queue but the service time needed by the customers in the queue may be longer thus
resulting in an even longer waiting time. Another side effect is that it can lead to disorder
during the crowded hours and the most recent incident is the fight at McDonalds during
their Hello Kitty promotion.

POSSIBLE FURTHER STUDIES

The simulation of queuing system can be applied to many real- world applications
from car parking in multiple-storey carparks to ship docking at seaports. If it were
possible to improve the queues, there would be more profits made and more time to carry
out business than ever before, which would be very useful in this fast paced world.

ACKNOWLEDGEMENTS
Our appreciation and gratitude extends mostly to Dr. Tay Seng Chuan at NUS for
being a very helpful and a good-humoured mentor. Dr. Tay has put in time and effort to
teach us the discrete-event simulation technique, and has assisted us in collating the
statistical data and analyzing the simulation results. We would also like to thank our
school teacher Mr. Wong Hong Weng, for helping to provide assistance in keeping us
updated and making us meet our mentor regularly. Thanks also go to our parents for
supporting our mentorship.
REFERENCES
[1] D. Jerry Banks, John S. Carson, Discrete- Event System Simulation, Prentic Hall.
[2] K. Watkins, Discrete Event Simulation in C, McGraw-Hill.