© 1998 Pär Karlsson, Department of Telecommunications and Mathematics,

University of Karlskrona/Ronneby

Teletraffic theory

Laboratory exercise 3:

Queuing Theory

Name Program Email Box no. Approved

Introduction

1.1 Objectives

By studying a fictional communication system, further knowledge about queuing theory and
performance analysis should be gained. The theoretical emphasis is on M/G/1 theory and priority
queues.

1.2 Prerequisites

To be able to follow this exercise you are required to do the following before the exercise starts:

• Performing all tasks marked “home exercises” (will be checked by a small test before the exercise
can commence).
• Study the relevant material in the textbook and presented on the lectures.
• Prepare for the exercise by looking through the tasks to be executed during the exercise.

1.3 Examination

To complete the exercise successfully you are required to:

• Participate in the exercise at the scheduled time (the laboratory exercises should be performed at
this time, not earlier).
• Hand in this manual, completed with thorough answers to all exercises.
• Write your answers in a pretty, readable manner.

2 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

2 Home exercises
2.1 System description

During this exercise, we will study the following, imaginary, communication system. The system,
which is packet based, transports two types of traffics, each with its own properties and requirements.
We will for the remainder of this exercise focus on a single node in this communication system, see
Figure 1. In order to keep things simple we assume that all packets arriving to the node are destined
for the same outgoing link. The bandwidth of the outgoing link is denoted B Mbit/s. Users that
generate traffic are connected to this node with access links that have a much higher bandwidth than B.
The total number of users is M.

Communication node
1
2

M Outgoing link
Access links

Figure 1: System overview

The first type of traffic is a data communication service. This is a non real-time service so it can
tolerate rather large delays of individual packets. Hereafter this service is denoted "data". Assume that
each user generates data packets according to a Poisson process with intensity λd. (Note: this
assumption is probably somewhat unrealistic.) Including lower layer headers, each data PDU (Protocol
Data Unit) is ld bytes (octets) long.

The second type of traffic comes from a voice communication application that has stringent delay
requirements. Short samples of voice are packetized and sent individually as packets over the system.
Voice packets from each user arrive according to a Poisson process with intensity λv. (Again, this is
somewhat unrealistic.) The size of a voice PDU is lv bytes.

In front of the outgoing link, buffering of packets is needed, for our analysis we can assume this buffer
to be of unlimited capacity.

During the exercise we will compare two different solutions to the internal structure of the node. The
first, and simplest, approach treats the data and voice packets identically. As packets arrive to the
node, they are placed in a buffer. The packets are then served in FIFO order, see Figure 2.

Data packet

Voice packet

Data and
voice buffer
Figure 2: Design with common buffer

3 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

2.1.1
Explain why the data and the voice service differ in their delay requirements.

2.1.2
Explain why a buffer is needed in front of the outgoing link.

2.1.3
What advantages and disadvantages can you think of with the suggested design?

2.1.4
Given the definitions above, determine the time it takes to transmit a data and a voice PDU over the
outgoing link. Call these times xd and xv, respectively.

xd xv

2.1.5
Determine the probability that an arriving packet is a data or a voice packet.

pd = P(data packet) pv = P(voice packet)

4 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

2.2 M/G/1

Let us now consider a theoretical modeling of the system described above. What we mainly are
interested in is the queuing process in the buffer.

2.2.1
What is the arrival process of packets into the node?

2.2.2
What is the service time distribution for sending packets over the outgoing link?

2.2.3
Determine the first and second moment of the service time distribution.

x x2

2.2.4
2
Determine C x , the squared coefficient of variation of the service time distribution.

5 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

2.2.5
Considering your answers to 2.2.1 and 2.2.2, suggest a queuing system that can be used to model the
system.

2.2.6
Determine ρ - the utilization of the system, as a function of M, λd, λv, ld, lv, and B.

2.2.7
Use your answers to 2.2.1 - 2.2.6 to derive the mean number of packets in the node expressed as a
function of M, λd, λv, ld, lv, and B.

2.2.8
Determine the mean waiting time in the buffer for an arbitrary packet as a function of M, λd, λv, ld, lv,
and B.

6 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

2.3 Priority queuing

Data buffer

Voice buffer

Figure 3: Design with separated buffers

Here data and voice packets are separated and placed in different buffers. Once the transmission of a
packet has finished over the outgoing link a new packet is fetched from the voice buffer and
transmitted. If no voice packet is available, a data packet is transmitted. If no packet is available at all
in the system, the first packet to arrive of any type will receive service. The packets are served in a
non-preemptive manner.

2.3.1
Compared to the first design, what advantages/disadvantages does this design display?

2.3.2
Explain the meaning of “a non-preemptive manner” in the context of the system described above.

2.3.3
Explain why a preemptive service of packets (as opposed to a non-preemptive) is less desirable in the
case of transmission of packets.

7 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

2.3.4
Considering the design presented above, suggest a queuing system that can be used to analyze the
performance.

2.3.5
Determine the first and second moments of the service time distributions for voice and data packets.

xv xd xv2 xd2

2.3.6
Determine ρv and ρd, the offered load of data and voice traffic as a function of M, λd, λv, ld, lv, and B.

2.3.7
Determine the mean delays caused by buffering for packets belonging to both services (voice and data
packets) as functions of M, λd, λv, ld, lv, and B.

8 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

3 Laboratory Exercises

3.1 M/G/1

Using your results from the home exercises, we will now further study the performance of our
proposed system. Unless otherwise stated, use the following parameters in the investigations below.

B 10 Mbit/s
λv 10/s
λd 5/s
lv 100 bytes
ld 1500 bytes

3.1.1
Complete the following table.

xv

xd

P(voice packet)

P(data packet)

C x2

3.1.2
Considering design approach number one, determine the maximum number of users (Mmax) that can be
connected to the node before it becomes overloaded.

9 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

3.1.3
Would you recommend connecting Mmax users to the node? What can be said about the expected
waiting times?

3.1.4
Using your result from 2.2.8, write a Matlab function that returns the average buffering time as a
function of M, λd, λv, ld, lv, and B:

IXQFWLRQ : ZGHVLJQ 0 ODPEGDG ODPEGDY OG OY % 

where : and 0 are vectors, and ODPEGDG, ODPEGDY, OG, OY, and % are scalars.

Test case:
ª ZGHVLJQ>  @     H

DQV

  

3.1.5
Using your function, plot the mean buffering time as a function of the number of users. (Attach the
plot as an appendix to this report). (Hints: a good idea is to use a logarithmic scale for the y-axis, see
the Matlab command VHPLORJ\. To annotate your plot you can use [ODEHO, \ODEHO, WLWOH.)

3.1.6
Let us assume that the maximum tolerable mean buffering delay for the voice packets is 0.5 ms. What
is the maximum number of users now?

10 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

3.1.7
Let us consider what happens if the data PDU size (ld) is decreased 10 times and λd is increased 10
times. This corresponds to keeping the offered traffic from each user constant. Note: in real life this
would probably lead to an increase in the traffic since we usually have a fixed amount of overhead in
each PDU.

What do you expect to happen with the average buffering delay with this new setup (keeping all other
parameters constant)? Explain why. (Hint: calculate C x2 .)

3.1.8
Using Matlab, plot the mean buffering time as a function of the number of users. (Attach the plot as an
appendix to this report).

3.1.9
Let us again assume that the maximum tolerable mean buffering delay for the voice packets is 0.5 ms.
What is the maximum number of users now?

11 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

3.2 Priority queuing

3.2.1
Considering the second design approach, determine the maximum number of users that can be
connected to the node before the queue of voice packets grows unbounded?

3.2.2
How is then the situation for in the data buffer? What is the maximum number of users if both the
voice and data queue should be stable?

3.2.3
Using your result from 2.2.8, write a Matlab function that returns the average buffering times for data
and voice packets as a function of M, λd, λv, ld, lv, and B:

IXQFWLRQ >:Y :G@ ZGHVLJQ 0 ODPEGDG ODPEGDY OG OY % 

ZKHUH :Y :G DQG 0 DUH YHFWRUV DQG ODPEGDG ODPEGDY OG OY DQG % DUH VFDODUV

Test case:

ª >ZYZG@ ZGHVLJQ>  @     H

ZY

  

ZG

  

3.2.4
Using your function, plot the mean buffering times for data and voice packets as a function of the
number of users. (Attach the plot as an appendix to this report).

12 (13)
 Teletraffic theory, Laboratory exercise 3: Queuing Theory, © ITM, HK/R, 1998

3.2.5
Let us assume that the maximum tolerable mean buffering delay for the voice packets is 0.5 ms. What
is the maximum number of users now? How does your answer compare to the answer in 3.1.6?

3.2.6
The average delay an arbitrary packet (voice or data) experience can be defined in different ways. Plot
the average delay (as a function of the number of users) produced by weighting the individual delays
on voice and data packets with pv, pd or ρv / (ρv+ρd), ρd / (ρv+ρd). Comment on the result. (Hint:
compare the plots with the one produced in 3.1.5).

3.2.7
Our study so far has only been concerned with the mean delay caused by buffering. What other
parameters, besides the mean, can you think of that might be of importance?

13 (13)