A formal perspective on the view selection problem

Abstract

The view selection problem is to choose a set of views to materialize over a database schema, such that the cost of evaluating a set of workload queries is minimized and such that the views fit into a pre-specified storage constraint. The two main applications of the view selection problem are materializing views in a database to speed up query processing, and selecting views to materialize in a data warehouse to answer decision support queries. We describe several fundamental results concerning the view selection problem. We consider the problem for views and workloads that consist of equality-selection, project and join queries, and show that the complexity of the problem depends crucially on the quality of the estimates that a query optimizer has on the size of the views it is considering to materialize. When a query optimizer has good estimates of the sizes of the views, we show that an optimal choice of views may involve a number of views that is exponential in the size of the database schema. On the other hand, when an optimizer uses standard estimation heuristics, we show that the number of necessary views and the expression size of each view are polynomially bounded.