Peer-to-peer Web caching: hype or reality?
Yonggen Mao, Zhaoming Zhu, Weisong Shi · 2004
Peer-to-peer Web caching has attracted a great attention from the research community recently, who believes it as a potential killer application for peer-to-peer networking. However, the observed results from several previous efforts are not consistent, even controversial. In this paper, we systematically examine the design space of peer-to-peer Web caching systems in three orthogonal dimensions: the caching algorithm, the document lookup algorithm, and the peer granularity. Based on the observation that traditional URL-based caching algorithm suffers considerably from the fact of cacheability decrease caused by the fast growing of dynamic and personalized Web content, we propose to use the digest-based caching algorithm. In addition to compare two existing document lookup algorithms, we propose a simple and effective geographic-based document lookup algorithm. Four different peer granularities, i.e., host level, organization level, building level, and centralized, are studied and evaluated using a seven-day Web trace collected at a medium-size education institution. Using a trace-driven simulation, we compared and evaluated all design choices in terms of two performance metrics: hit ratio and latency reduction. The experimental results suggest that: (1) ideally, the digest-based caching algorithm could improve the cacheability of Web objects substantially, from 6.9 % (URL-based) to 62.0 % (digest-based); (2) the document sharing among peers is very effective, from 22.0 % (building level) to 34.2 % (host level); (3) the average user-perceived latency is reduced three to six times compared with the measured latency at all peer granularities using the hierarchical index-based (home1 in our jargon) document lookup algorithm; (4) the proposed geographic-based document lookup algorithm has comparable hit ratio and significant latency reduction. Finally, several implications derived from these observations are also listed.