A dynamic view materialization scheme for sequences of query and update statements

Abstract

In a data warehouse design context, a set of views is selected for materialization in order to improve the overall performance of a given workload. Typically, the workload is a set of queries and updates. In many applications, the workload statements come in a fixed order. This scenario provides additional opportunities for optimization. Further, it modifies the view selection problem to one where views are materialized dynamically during the workload statement execution and dropped later to free space and prevent unnecessary maintenance overhead. We address the problem of dynamically selecting and dropping views when the input is a sequence of statements in order to minimize their overall execution cost under a space constraint. We model the problem as a shortest path problem in directed acyclic graphs. We then provide a heuristic algorithm that combines the process of finding the candidate set of views and the process of deciding when to create and drop materialized views during the execution of the statements in the workload. Our experimental results show that our approach performs better than previous static and dynamic approaches. © Springer-Verlag Berlin Heidelberg 2007.