Predictions for web prefetching using recursive least squares filter

The rapid growth of Internet traffic has made bandwidth a scarce resource. Prefetching techniques in existence are either computationally heavy that could be run only on high end servers or waste bandwidth due to bad predictive performance. The Predictive Prefetching system we propose here using rapid converging Recursive Least Squares (RLS) filter algorithms is not only light on the server but also shows good accuracy in predicting web pages. Moreover, because of fast convergence of the algorithm, one starts experiencing reduced latencies soon after it has been put in place. Access patterns are studied from access logs maintained by the server. Firstly, these logs are subject to conventional volume construction techniques that. group websites into clusters of URL's which have high likelihoods of being accessed together. Once this has been done, the filter algorithm aids in effectively predicting future accesses. With high prediction accuracy, there is significant reduction of user-perceived latency and bandwidth consumption. We have obtained results from several logs. The results show that it is possible to make accurate predictions that give the percentage of prefetched files accessed of more than 40%.