Dynamic Construction and Administration of the Workload Graph for Materialized Views Selection

To offer the best performance to each application (e.g., decision support systems) administration and optimization of database management systems is needed. In order to reduce administration costs and to provide continuous adaptation to changing workload patterns, selfmanagement techniques have been in the focus of researchers and DBMS vendors for recent years. One important topic in this area is the selection of materialized views, i.e., complex queries of OLAP systems benefit from pre-computed results. In this work, we present an automatic selection of materialized views using statistics of a workload which is represented in the query graph model. We propose merging of queries of a workload into a workload graph and the computation of merged views. Hence, we introduce a cost model to select the views to be materialized. The proposed techniques are evaluated using the TPC-H benchmark and compared to an industrial design advisor. Furthermore, our approach can be used in static as well as in on-line self-tuning.