245

1 Introduction

A test platform is required for Quality of Service

testing in IP based networks that carry all types

of services. The services that are provided can

e.g. include Internet access, home office, tele-

phony, video and TV distribution. These ser-

vices have different traffic characteristics and

Quality of Service requirements and the IP net-

work will therefore require some support for dif-

ferentiated QoS provisioning. Such a test plat-

form should include generation of realistic traf-

fic and monitor functions that are able to study

the details of traffic streams in the network. This

paper presents a flexible test platform currently

being used for QoS performance testing in an

experimental, IP based, communication platform

that will provide differentiated services.

For the purpose of QoS testing of new applica-

tions and network mechanisms in IP networks,

a generator of controllable, re-producible, scal-

able, synthetic but realistic traffic is required.

This is the motivation for the development of

GenSyn – a generator of synthetic IP traffic

implemented in Java. The generator will typi-

cally produce traffic in a controlled testbed en-

vironment where there are few real users and a

corresponding low traffic load. Realistic, con-

trollable, and reproducible background traffic

load is of great importance in order to test the

QoS of new applications, i.e. interactive video.

In addition, to increase the understanding and

get experience with the configuration of new

QoS network mechanisms the network must

be offered a high load of heterogeneous traffic

mixture.

Besides generating traffic, a set of performance

parameters must be monitored to study the ser-

vice performance defined at IP layer or transport

layer (UDP/TCP), or even higher layers, e.g. for

real-time services the parameters can be IP

packet delay, jitter and loss, while the quality of

TCP connections can be measured as throughput

at TCP or IP layers. Other performance charac-

teristics can be derived based on these parame-

ters, e.g. the service availability can be defined

as the state where the end-to-end delay is less

than a specified limit, and the measurement

accumulates the fraction of the time the system

is in this state. A flexible measurement set-up

has been achieved by deploying PCs dedicated

to passively monitor traffic using specialized

hardware, so-called DAG PCI cards. However,

the specification of a measurement and monitor-

ing platform depends on what performance

parameters that are to be observed.

A Distributed Test Environment for IP

Performance Evaluation

POUL E. HEEGAARD AND BRYNJAR Å. VIKEN

As input to IP Traffic Engineering it is required to conduct measurement both on live networks and test

networks. This paper presents a distributed test platform currently being used for QoS performance test-

ing in an experimental, IP based communication platform that will provide differentiated services. The

measurement platform developed for QoS tests consists of two main components; (i) DAG monitors to

derive performance measures like end-to-end packet loss and one-way packet delay accurately, and (ii)

GenSyn, a Java-based traffic generator running on dedicated PCs. This testbed configuration is a very

flexible platform that opens for doing many exciting and controlled QoS performance evaluation mea-

surements in an IP network.

Most testbeds have no, or very low traffic load, so traffic generators are required. The traffic mixture that

is generated by GenSyn is controllable, reproducible, synthetic traffic according to an aggregation of

stochastic application models. Currently available are models of web and FTP clients that generate TCP

traffic by downloading pages and files from actual web servers, and models that generate UDP traffic

from a video server (using MPEG), from voice over IP (VoIP), and in a Constant Bit Rate (CBR) stream.

Besides generating traffic, there is a need for accurate traffic measurements in order to derive perfor-

mance measures like unidirectional end-to-end packet loss and delay. This has been achieved by

deploying PCs dedicated to traffic monitoring using specialized hardware, so-called DAG PCI cards.

For the purpose of QoS testing of new applications and network mechanisms in IP networks, several

test scenarios are defined. The scenarios specify the application mixture (VoIP, VoD, FTP, web, TV,

etc.) and/or protocol mixture (TCP, UDP), load level, traffic matrix, and network configuration (DiffServ,

Best Effort etc.).

The experiences from running experiments in such a distributed test environment revealed a need for a

support system assisting the analysts in setting up the experiments, collecting data, and post process-

ing it. Such support functions are under development and are briefly presented in this paper.

Brynjar Å. Viken (31) is
Research Scientist at Telenor
Research and Development,
Trondheim. His research inter-
ests are performance measure-
ments and analysis of communi-
cation networks with a special
interest in IP networks. He is
currently pursuing a PhD at the
Norwegian University of Science
and Technology.

[email protected]

Poul E. Heegaard (37) is Senior
Research Scientist at Telenor
R&D, Trondheim. His research
interests are within the areas of
traffic and dependability evalua-
tion of telecommunication sys-
tems. He has a special interest
in speedup simulation tech-
niques for assessment of sys-
tems with rare events, and moni-
toring and measurements of IP
networks. He received his Mas-
ter’s degree in 1989 and his
Dr.Ing (PhD) in 1998 in Telemat-
ics from the Norwegian Univer-
sity of Science and Technology
(NTNU). Heegaard holds an
adjunct (20 %) associate profes-
sorship in simulation at the
Department of Telematics,
NTNU. From 1989 to 1999 he
was research scientist at SIN-
TEF Telecom and Informatics.

[email protected]

Telektronikk 2/3.2001