Performance Analysis of View Materialization Strategies for Select-Project-JoinExpressions

摘要

In conventional relational database systems, a view is a virtual relation whosedefinition is stored in the systems catalog. When a query is issued on the view, the system retrieves the view from the catalog and modifies the query to an equivalent one on the base relations. Recently several approaches have been proposed that store some form of the computed view as a method for improving the performance of queries on relational database. This thesis develops a computer program to empirically compare and evaluate three view materialization strategies: query modification, semi-materialization and full materialization. The program simulates user updates and queries, and measures the cost performance of the three materialization strategies. The strategies are compared for select-project-join expressions under three different view models. The results show that the most efficient view strategy is heavily application dependent. The performance of semi-materialization and full materialization, however, are comparable for most conditions tested, and preferred over the conventional query modification method.