Conclusions

  In this paper we present a Top-10 approach for prefetching World Wide Web documents. Top-10 prefetches only the most popular documents (that is where the name comes from) and only to clients that will be able to use them. We use trace-driven simulations of server traces to evaluate the costs and benefits of our approach. Based on our experimental observations we conclude:

• The Top-10 approach to prefetching may result in significant performance improvements. Our experimental results suggest that Top-10 manages to prefetch (up to) 60% of future requests, with little (less than 20%) corresponding increase in traffic (see fig. 10).
• Top-10 has shown to be robust over a wide variety of parameters and server loads. We have used traces from 5 different servers both in Europe and the States. Top-10 always managed to result in good performance for all traces without significant traffic increase.
• Prefetching the most popular documents is a simple but effective prefetch heuristic. It requires very little effort to be computed on the server side, while it provides performance comparable to (if not better than) previously proposed sophisticated prefetching heuristics.