Improving World-Wide-Web performance using domain-top approach to prefetching

The exponential rate of growth of the World Wide Web has led to an increase in Internet traffic, as well as a serious degradation in user-perceived latency while accessing "Web pages". One way to reduce the latency is through the use of caching. Prefetching method may be applied to further increase the cache hit ratio, by anticipating and prefetching future client requests. The authors propose a domain-top approach to prefetching which combines the proxy's active knowledge of the most popular domains and documents with client access profiles. Our goal is to increase the hit ratio by proxy prefetching and to put a little burden on the proxy and the network. In our scheme, proxy finds the popular domains using access profiles and searches the popular documents in each domain. Based on these Top-Domain and Top-Documents, proxy makes the rank list for prefetching, the client requests a file in a certain domain and proxy forwards to them their most popular documents in the rank list. Finding the popular domains and documents does not require heavy computation power to the proxy, but only needs a very small amount of rank list that stores them at the proxy. This approach can be implemented without changes to server and client. We did trace driven simulation by using the access logs from the KAIST proxy server to evaluate domain-top prefetching. We demonstrate the effectiveness of the domain-top approach by evaluating various prediction table constructions across a collection of large proxy logs. The results show that the domain-top prediction algorithm can raise hit ratios up to 20% in worst cases, and that domain-top can increase the hit ratio up to 450% in the best case.