Related Work

Studies on techniques which aim to reduce the latency of Web accesses include LowLat [19] and WebExpress [13]. LowLat differs from our system in that it requires a process to be located near the Web server. WebExpress differs in that it uses file caching, forms differencing, protocol reduction, and the elimination of redundant HTTP header transmission to reduce the bandwidth used. Furthermore, WebExpress multiplexes multiple HTTP requests over a single link to reduce the TCP setup overhead. Currently, our system transmits HTTP documents through standard HTTP/1.0.

Studies into speculative pre-fetch of Web documents include work done by the OCEAN group [6,9,8], ICS-FORTH [15], Tenet [17,18], and Wcol [5]. OCEAN's approach differs in that they use both server initiated pre-fetch as well as client-initiated pre-fetch. Further, they use a Random Walk User Model and a DSP User Model to model usage patterns. ICS-FORTH differs in that they employ a server initiated pre-fetch with the help of a Top-10 Approach. Tenet represents usage pattern on the server through dependency graphs. Similar to our pre-fetch with notification, their server makes the predictions and the client initiates the pre-fetches. Wcol differs from our profile-based pre-fetch in that they parse the HTML files and pre-fetch both the links and the inline images. Wachsberg [20] describes the use of a model similar to ours. A commercial product that does speculative pre-fetch is PeakJet [3].

Studies on geographical push caching [11,12] by the VINO research group involves server initiated pushing and differs from our client initiated approach.

Studies into collaborative data filtering include Tapestry [10], and FIREFLY [1]. JunkBusters [2] is a proxy server that also filters HTTP requests. Our work is similar to the architecture that Zenel describes for intelligent filtering in low-bandwidth environment in that we make use of an intermediary (proxy).