Preprocessor of Materialized Views Selection

Abstract

The availability and performance of data warehouse is gradually degrading with variable requirements. The set of materialized views is far from being optimal, so it is necessary to implement the dynamic adjustment to meet the demand of the users. Since the current static algorithms are not suitable for this purpose due to their larger space search and higher time consumption, this paper proposes PMVS (Preprocessor of Materialized Views Selection), which is composed of three algorithms: QSDM (Query Set Dynamic Management), CVLC (Candidate View Lattice Construction) and CVF (Candidate View Filter). Of the three algorithms, QSDM monitors the distribution of each query and determines by hypothesis testing whether the query should be added into or discarded from the query set. Based on the given query set, CVLC is in charge of producing candidate view set, which has been proven to be sufficient and necessary for selecting the best set of materialized views. As a heuristic algorithm, CVF utilizes the data sparse in multi-dimensional datasets to remove a part of candidate views, with very limited contribution to the optimal solution. The contrastive experiment indicates that PMVS can be employed by the static algorithms to reduce the amount of previous views effectively. The cost of static algorithms on space and time can be decreased to fit online demand.