Introduction

Abstract

Integrated broadband networks are expected to support various traffic types such as data, voice, image and video. Traffic generated from these services is substantially different in its statistical characteristics and the QoS that the network requires to maintain throughout the session of the call as well as during the call set - up phase. For example, voice traffic has a bandwidth requirement of several Kbps and is delay sensitive, while high speed data traffic used for file transfers or LAN/WAN interconnection is hundreds of Mbps and is loss sensitive [5]. The complexity of broadband traffic requires modeling and analysis which can be quite unconventional in the engineering sense. Traditional modeling tools and techniques, both theoretical and empirical, have been able to characterize and understand the behavior of broadband traffic to a rather limited extent. However, the discovery of scaling in the measured teletraffic has led to modeling solutions that can approximate the data characteristics much better than previous techniques. The statistics of scaling behavior presents many challenges, the key one being, non - stationary behavior: This can be handled by exploiting ideas from computer networks, signal processing, information theory, dynamical systems and thermodynamics [6].