Materialized View Selection Using Iterative Improvement

Abstract

A data warehouse is designed for the purpose of answering analytical queries, posed on them for decision making. The complex and exploratory nature of analytical queries which, when processed against large historical information in the data warehouse, consume a lot of time for processing. As a result, the query response time is high. Materialized views provide an alternative platform to address this problem of poor query response time. These views store aggregated and summarized information separately from a data warehouse with the aim of answering analytical queries. All views cannot be materialized, as the number of views is exponential in respect of number of dimensions. Also, optimal view selection is an NP-Complete Problem. Several view selection algorithms exist with most selecting views empirically or based on heuristics like greedy or evolutionary. In this paper, an algorithm based on iterative improvement, a randomized search heuristic technique for selecting top-K views for materialization is proposed. It is shown that the proposed algorithm, in comparison to a well known greedy algorithm, is able to select comparatively better quality views for higher dimensional data sets.