Performance tuning of large-scale distributed WWW caches

World Wide Web (WWW) caches, such as the Harvest cache, its successor Squid, and a cache proposed by Malpani, Lorch, and Berger (1996) from the University of California, Berkeley (referred to hereafter as the Berkeley cache) implement proxy to proxy communication. This communication mechanism unifies the communicating proxy caches and makes them perform as a single cache that is distributed over multiple proxy machines. In this paper, we investigate how the size of the distributed cache affects its performance. We compare the performance results for two different distributed caches: Squid and Berkeley. We show how to choose the number of proxy machines in a distributed cache in order to avoid network congestion and prevent overloading of the proxy machines.