Caching strategies for dataintensive web sites

Data-intensive Web sites serve large volumes of pages whose content is dynamically extracted from a database. Such Web sites have very high software development and maintenance costs and in general oer poor response times due to the heavy interaction with the database system. This paper introduces the Weave management system de-veloped at INRIA, which alleviates the above shortcomings of data-intensive Web sites. Weave relies on the declara-tive specication of Web sites and oers a number of tools for the easy implementation, deployment and monitoring of the specied site. Weave features a customizable cache system that implements the optimal data materialization strategy according to the Web site's specics: it can cache database data, XML fragments and HTML les. To ex-plore Weave's performance we have built a Web site based on the TPC/D benchmark database using the WeaveBench test platform. We conducted a number of experiments with various data materialization strategies supported by our system. Results clearly show that in the general case, a mix of dierent caching policies is required to achieve op-timal performance.