Warehousing and Incremental Evaluation for Web Site

Recently, several systems have been proposed for building Web sites from declarative spec-iications. The salient feature of these systems is that they model both the structure and the content of a Web site as a view over some existing raw data. In this \Web site as view" paradigm, a critical issue is when to compute the site. One approach is to materialize the site completely before the users browse it: the disadvantages of this approach are high maintenance costs and stale data. Moreover, this approach is not applicable if the Web site has forms, because then the queries deening the Web site depends on user input. The other extreme approach is to precompute only the root(s) of the Web site, and then compute dynamically (at \click time") the queries that retrieves the information required to display the next page. In this paper we address the problem of choosing the optimal connguration for run-time management of a declaratively speciied Web site, trading oo the beneets of precomputation and incremental evaluation. First, we identify several novel optimizations that need to be considered in addition to optimizations obtained by precomputing views. These optimizations include simplifying queries based on contexts that can be derived from the browsing patterns, and lookahead computation. Second, we describe an algorithm for cost based search of a space of conngurations which include the above optimizations. Finally, given a connguration we show how to reformulate our Web-site deenition to exploit the views and data being cached.