Author Description

This study investigates the application of a clustering technique in a Web Prefetching approach that uses the Prediction by Partial Match (PPM) algorithm. The clustering method presented herein is based on the Partitioning Around Medoids algorithm. Past study [PM99] shows that Web servers can benefit from the implementation of a PPM Web Prefetching algorithm. This study changes the experiment target to the proxy server. The prediction engine is moved to the proxy side. Web proxy trace files are used to execute simulations on the new system. The results indicate that the performance of the Web Prefetching system is improved significantly by the client clustering process. The simulation suggests that certain groups of clients are able to enjoy the advantages of employing client clustering. The clustered prediction models are effective in situations where there are clear clusters of customers who share similar web access patterns.