Semi-materialization: a performance analysis

The performance of queries on relational views can be greatly improved by keeping a stored copy of the materialized view. However, the cost of maintaining materialized views during updates on base relations is often high. This is especially so in the case of view definitions with universal quantifiers. An alternative approach called semi-materialization has been proposed, whereby carefully chosen redundant subsets of data of individual relations are stored. These subsets represent an intermediate state of view evaluation rather than a complete evaluation. The authors compare the performance of queries over views using query modification, semi-materialization and full materialization. The view definition expressions considered are: select-project-join, and a general form of calculus expressions. The results show comparable performance of full-materialization and semi-materialization for select-project-join expressions, and superior performance of semi-materialization for general expressions