Modeling View Selection as a Constraint Satisfaction Problem

Using materialized views can highly speed up the query processing time. This paper deals with the view selection issue, which consists in finding a set of views to materialize that minimizes the expected cost of evaluating the query workload, given a limited amount of resource such as total view maintenance cost and/or storage space. However, the solution space is huge since it entails a large number of possible combinations of views. For this matter, we have designed a solution involving constraint programming, which has proven to be a powerful approach for modeling and solving combinatorial problems. The efficiency of our method is evaluated using workloads consisting of queries over the schema of the TPC-H benchmark. We show experimentally that our approach provides an improvement in the solution quality (i.e., the quality of the obtained set of materialized views) in term of cost saving compared to genetic algorithm in limited time. Furthermore, our approach scales well with the query workload size.