Static caching in Web servers

Most existing studies on Web document caching consider Web proxies. The focus of this paper is caching in primary Web servers. This type of caching decreases request response time and server load. We use log les from four Web servers to analyze the performance of various proposed cache policies for Web servers: LRU-Threshold, LFU, LRU-SIZE, LRU-MIN, and the Pitkow/Recker policy. We also study application of LRU-k, a caching policy initially designed for database systems, to Web servers. Augmented LRU-k with a threshold parameter makes it an eecient caching policy for Web servers. Web document access patterns change very slowly. Documents, that were popular today, will stay popular tomorrow. Based on this fact, we propose and evaluate static caching, a novel cache policy for Web servers. In static caching, the set of documents kept in the cache is determined periodically by analyzing the request log le for the previous period. The cache is lled with documents to maximize cache performance provided document access patterns do not change. The set of cached documents remains constant during the period. Surprisingly, this simple policy results in high cache performance, especially for small cache sizes. Finally, we discuss implementation issues of the studied caching policies. We show that many policies incur signiicant CPU overhead and memory fragmentation. On the contrary, static caching does not have these problems.