Abstract

In this paper, we study prefetch techniques in the WWW, in which we predict which files will be needed in the near future and download some of them before they are requested by the user. Our prefetch scheme includes two algorithms: the prediction algorithm and the threshold algorithm. The prediction algorithm estimates the probability with which each file will be requested in the near future. The threshold algorithm computes the prefetch threshold for each server. An important contribution of this paper is a formula we derived to determine the prefetch threshold dynamically based on the system load, capacity and the cost of time and system resources to the user. Simulations driven by trace files show that using access information from the client can achieve high successful prediction rates, while using that from the server can result in more hits in general. We have also developed a prefetch program at the client site which assists users in browsing faster and more efficiently. 1 Intr...

Citations