Hash routing for collections of shared Web caches

• Published in: IEEE network Vol. 11; no. 6; pp. 37 - 44
• Main Author: Ross, K.W.
• Format: Journal Article
• Language: English
• Published: New York IEEE 01-11-1997; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Cache; Cache memory; Cache storage; Delay; Measurement; Performance analysis; Robustness; Routing; Routing protocols; Servers; Uniform resource locators; Web pages; Web server; World Wide Web
• ISSN: 0890-8044; 1558-156X
• DOI: 10.1109/65.642358

Shared Web caches, also referred to as proxy Web servers, allow multiple clients to quickly access a pool of popular Web pages. An organization that provides shared caching to its Web clients will typically have a collection of shared caches rather than just one. For collections of shared caches, it is desirable to coordinate the caches so that all cached pages in the collection are shared among the organization's clients. In this article we investigate two classes of protocols for coordinating a collection of shared caches: the ICP protocol, which has caches ping each other to locate a cached object; and the hash routing protocols, which place objects in the shared caches as a function of the objects' URLs. Our contribution is twofold. First, we compare the performance of the protocols with respect to cache-server overhead and object retrieval latency; for a collection of shared caches, our analysis shows that the hash-routing schemes have significant performance advantages over ICP for both of the performance metrics. The existing hash-routing protocols assume that the cache servers are homogeneous in storage capacity and processing capability, even though most collections of cache servers are vastly heterogeneous. Our second contribution is to extend a robust hash-routing scheme so that it balances requests among the caches according to any desired distribution; the extended hash-routing scheme is robust in the face of cache failures, is tunable for heterogeneous caches, and can have significant performance advantages over ICP.