Multi-dimensional declustering methods for parallel database systems

Abstract

This paper describes and evaluates new methods for relation declustering in parallel databases. To process queries in parallel, relations are partitioned across multiple processors, typically by using the value of one single attribute. This kind of declustering has resulted in poor performance in the presence of data skew. Alternatively, the work contained herein proposes several strategies to decluster a relation through the use of multiple attributes. To demonstrate the validity of our approach, a thorough performance evaluation is done. The findings demonstrate the effectiveness of this kind of partitioning methods as opposed to traditional ones. In addition we analyze the performance of the different strategies relative to the speed-up and scale-up metrics. To sum up, performance results reveal that multi-dimensional declustering methods constitute a very promising alternative to conventional one-dimensional methods to partition relations in parallel database systems.