Evaluating Server - Assisted Cache

1 To reduce user-perceived latency in retrieving documents on the world wide web, a commonly used technique is caching both at the client's browser and more gainfully (due to sharing) at a proxy. The eeec-tiveness of Web caching hinges on the replacement policy that determines the relative value of caching diierent objects. An important component of such policy is to predict next-request times. We propose a caching policy utilizing per-resource statistics of inter-request times. Such statistics can be collected either locally or at the server, and piggybacked to the proxy. Using various Web server logs, we compared existing cache replacement policies with our server-assisted schemes. The experiments show that utilizing the server knowledge of access patterns can greatly improve the eeectiveness of proxy caches. Our experimental evaluation uses an interval caching model. Interval caching values the utility of a unit of cache storage as a function of time. Instead of the usual trade-oos of proot (combined value of cache hits) and cache size we measure tradeoos of proot and caching cost (average cache size). Interval-caching policies have analogous cache replacement policies, and policies and their relative performance in the two models are closely tied. Interval-caching allowed us to evaluate and compare diierent replacement policies by using server logs, without having to construct a full workload model for each client's cache.