Statistical Characterization of a World Wide Web Browsing Session

Abstract

The growing availability of Internet access has led to a sharp increase in the use of the World Wide Web(WWW). In the near future, a significant component of the Internet traffic into and out of the home is expected to be WWW related traffic. If we are to design broadband access systems to carry this traffic, it is important to be able to model it accurately, with the hope that we can track and predict the impact of this relatively new type of traffic. The upstream packets presented by a subscriber engaged in a web browsing session, cannot be modelled as an independant one-way source since the actual inter-arrival between the packets depends on the feedback mechanisms and restrictions imposed by the transport/application layer protocols. We need to model the complete packet exchange between the WWW Client/Server to generate realistic source traffic for such systems. In this paper we report statistics on different parameters of the WWW browsing session obtained from actual WWW traces. We also present empirically derived analytical models for the different parameters . The statistical characterization of the different parameters directly lends itself to the development of a more realistic WWW-browser packet generator the basic details of which are addressed in this paper. This WWW traffic generator will be very useful in testing the performance of media access protocols and resource reservation algorithms in broadband access networks.