Telektronikk 2/3.2001

1 Introduction

The traditional Internet provides an unreliable

best-effort packet delivery. Packets can be lost,

errored, duplicated and delivered out-of-order.

The packet delivery is characterized by the

absence of service guarantees and fairness. New

emerging applications with the need for end-to-

end performance guarantees drive the introduc-

tion of service differentiation in IP networks.

Examples of these applications are IP telephony,

network games, audio and video streaming. The

introduction of service differentiation in IP net-

works increases the importance of traffic mea-

surements. Traffic measurements are vital for

several areas including network management,

traffic engineering, charging and billing, and

monitoring of service level agreements.

1.1 Actors Involved in Traffic

Measurements

Actors interested in collecting measurement data

are end-users, network providers, service and

content providers, vendors and researchers. The

respective actors have various perspectives and

needs of traffic measurements on various time-

scales. The time-scales may be classified as

short (seconds – minutes – hours), medium

(days – weeks) and long (months – years).

End-users need to gather measurement data to

ensure that the IP services they receive from

their service providers meet the agreed levels of

service. In the coming years as service differen-

tiation is introduced, traffic measurements will

play a major part in evaluating service quality

versus price for services an end-user receives.

The main focus of the end-user is end-to-end

performance of certain services where the net-

work is considered as a black box.

Network operators are focused on the network

domain under their management. Measurements

are important for the network operators both on

a day-to-day basis and for long term planning

and engineering. A network operator needs to

perform daily measurements to diagnose net-

work problems before they occur, perform trou-

bleshooting and solve network failures. Histori-

cal data form a foundation on which decisions

can be made. That is, planning, optimizing and

traffic engineering of the network domain. An

operator must also monitor performance to

ensure that the services delivered to peering net-

works, service and content providers, and end-

users meet the service level agreements. A net-

work operator will need information from peer-

ing networks in order to decide how to route

traffic of various service qualities. An end-to-

end path often crosses several network domains

and the end-to-end performance is beyond the

control of one individual network operator. Evi-

dently, efforts to both deliver and measure end-

to-end services require co-operation between

network operators, service providers and content

providers.

Service and content providers must rely on other

actors to provide end-to-end network services

for their customers. Thus, it is crucial for service

and content providers that their service level

agreements with other actors are fulfilled and

that the end-to-end services delivered satisfy the

QoS requirements of their customers. Service

and content providers can collect some measure-

ment data themselves, e.g. from service specific

equipment (Video on demand servers, VoIP

gateways, etc.), but depend on network

providers to measure network performance.

Traffic Measurements in IP Networks

BRYNJAR Å. VIKEN AND PEDER J. EMSTAD

There is an increasing need for traffic performance measurements in IP networks. The Internet has

become an important part of the communication infrastructure, and network operators and service

providers need to measure the flow of traffic. Today’s best effort networks give variable quality to users.

This may be alleviated in the near future offering differentiated services to the users. This article pre-

sents an overview of issues related to traffic measurements in IP networks, including characterization

of services, measures of interest, measurement methods, measurement infrastructures and post-pro-

cessing of measurement data. The main focus is on network-level performance measurements from

the perspective of a network operator. A novel model appropriate for precise and concise discussion

and definition of network level measurement issues is introduced and used to precisely describe net-

work-level performance metrics. A discussion of the implications on traffic measurements by introducing

means and mechanisms for service differentiation concludes the paper.

Dr.Ing. Peder J. Emstad (61) is

Professor at the Department of

Telematics at the Norwegian

University of Science and Tech-

nology. His research interests

are performance modelling and

analysis of communication net-

works and switching systems.

[email protected]

Brynjar Å. Viken (31) is

Research Scientist at Telenor

R&D, Trondheim. His research

interests are performance mea-

surements and analysis of com-

munication networks with a spe-

cial interest in IP networks. He is

currently pursuing a PhD at the

Norwegian University of Science

and Technology.

[email protected]

230