Parallel Maintenance of Materialized Views in Large-Scale Analytic Platforms

摘要

Speeding up query processing is an important issue in database management. Materialized views are largely used to address this issue. They have been proven successful for query performance optimization. However, updating data sources of the corresponding view requires maintaining the related views. Therefore, a view maintenance strategy is required. This paper presents a novel approach for materialized view maintenance that overcomes the limitations of prior approaches using a parallel "Divide and Conquer" strategy. The authors modeled the view maintenance problem using a new concept called "Multiple-Views-Matrix" as a matrix that brings up all affected views and their corresponding base relations. Moreover, the authors introduce a new algorithm for performing maintenance of multiple views, the proposed algorithm is able to use multiple parallelism and recursivity; this allows it to maintain multiple views and to process several updates at the same time. The authors show that this method provides a significant improvement in terms of maintenance process cost.