A group replacement algorithm for web caches

Abstract

With the immense growth of the Internet, Web caching has become an important area of research. By keeping a copy of the most popular files at the client side, caching is an effective mechanism to reduce network traffic and file retrieval time. In recent years several approaches have been made to improve cache performance, but none of them studied WWW traffic in detail. This thesis carries out an extensive analysis of WWW traffic, and based on its results, introduces a new caching policy that is based on a model of related URLs that can be thought of as a group. With the help of a trace driven cache simulation this policy is compared to the performance of conventional policies. The simulation results show that the policy improves cache performance, i.e. hit rate, by 10% to 20% depending on the size of the cache. Further, a comparison of the WWW traffic shows that the amount of bytes transferred is decreased by over 20%.