Materialized Views Selection in Distributed Networks

Abstract

Materialized views selection has recently received increasing attention in the database community. Although several greedy and heuristic algorithms for materialized views selection have already been proposed in centralized environments, their quality can not be guaranteed well in realistic distributed environments. Motivated by the above, under the constraint of maintenance and communication costs, this paper proposes two efficient algorithms in SPA (Super-Peer Architecture) distributed networks and uses the map/reduce distributed computation model to quickly find the optimal materialized views. For the first algorithm SMVSA, we first heuristically constructs the initial set of materialized views and then adjusts the set of materialized views based on simulated annealing. While for the second algorithm SMVMST, we reduce the materialized views selection to the problem of producing the minimum Steiner tree and obtain the approximate optimal set of materialized views. The detailed theoretical analyses and extensive experiments demonstrate that our proposed algorithms are both efficient and effective.