Taxonomy and design analysis for distributed Web caching

Proxy server caches reduce Internet traffic and improve response times. However, limited duplication among requests restricts most proxy server hit rates to under 50%, requiring other methods to expand cache sharing. The paper proposes a taxonomy for distributed network caching based on discovery, dissemination, and delivery methods. We analyze the match between taxonomy categories and Web characteristics, and use the taxonomy to classify current Web caching projects. Next we describe our cooperative Web caching protocol, in which proxy servers locate cache copies by looking in local metadata directories. Local lookup provides fast discovery, and allows multiple criteria to be used to select a cache site. We propagate metadata by lazy prefetching, in which returned objects carry metadata for related or popular objects. The protocol is simulated using empirically-derived analytical workloads. Results indicate that it substantially reduces server load and connection denials as compared to standard proxy server caching.