Markov Tree Prediction on Web Cache Prefetching

As Web accesses increase exponentially in the past decade, it is fundamentally important for Web servers to be able to minimize the latency and respond to users’ requests very quickly. One commonly used strategy is to “predict” what pages the user is likely to access in the near future so that the server can prefetch these pages and store them in a cache on the local machine, a Web proxy or a Web server. In this paper, we present an approach to effectively make page predictions and cache prefetching using Markov tree. Our method builds a Markov tree from a training data set that contains Web page access patterns of users, and make predictions for new page requests by searching the Markov tree. These predicted pages are prefetched from the server and stored in a cache, which is managed using the Least Recently Used replacement policy. Algorithms are proposed to handle different cases of cache prefetching. Simulation experiments were conducted with a real world data of a Web access log from the Internet Traffic Achieve and the results show the effectiveness of our algorithms.