A Behavioral Model of Web Traffic

Abstract--The growing importance of Web traffic on the Internet makes it important that we have accurate traffic models in order to plan and provision. In this paper we present a Web traffic model designed to assist in the evaluation and engineering of shared communications networks. Because the model is behavioral we can extrapolate the model to assess the effect of changes in protocols, the network or user behavior. The increasing complexity of Web traffic has required that we base our model on the notion of a Web-request, rather a Web page. A Web-request results in the retrieval of information that might consist of one or more Web pages. The parameters of our model are derived from an extensive trace of Web traffic. Webrequests are identified by analyzing not just the TCP header in the trace but also the HTTP headers. The effect of Web caching is incorporated into the model. The model is evaluated by comparing independent statistics from the model and from the trace. The reasons for differences between the model and the traces are given.