A performance analysis of alternative multi-attribute declustering strategies

During the past decade, parallel database systems have gained increased popularity due to their high performance, scalability and availability characteristics. With the predicted future database sizes and the complexity of queries, the scalability of these systems to hundreds and thousands of processors is essential for satisfying the projected demand. Several studies have repeatedly demonstrated that both the performance and scalability of a paralel database system is contingent on the physical layout of data across the processors of the system. If the data is not declustered properly, the execution of an operator might waste resources, reducing the overall processing capability of the system.

With earlier, single attribute declustering strategies, such as those found in Tandem, Teradata, Gamma, and Bubba parallel database systems, a selection query including a range predicate on any attribute other than the partitioning attribute must be sent to all processors containing tuples of the relation. By directing a query with minimal resource requirements to processors that contain no relevant tuples, the system wastes CPU cycles, communication bandwidth, and I/O bandwidth, reducing its overall processing capability. As a solution, several multi-attribute declustering strategies have been proposed. However, the performance of these declustering techniques have not previously been compared to one another nor with a single attribute partitioning strategy. This paper, compares the performance of Multi-Attribute GrId deClustering (MAGIC) strategy and Bubba's Extended Range Declustering (BERD) strategy with one another and with the range partitioning strategy. Our results indicate that MAGIC outperforms both range and BERD in all experiments conducted in this study.