A Dynamic and Multidimensional Declustering Method for Parallel Databases

The performance of parallel database systems is closely related to the declustering methods of databases. In the recent years, many declustering methods have been proposed. But most of them cannot support dynamic databases. A new dynamic and multidimensional declustering method called DCMD(dynamic coordinates module declustering) is proposed in this paper. DCMD not only supports dynamic databases but also has the advantages of multidimensional data declustering. DCMD consists of a component for initially declustering databases and a component for dynamically adjusting data distribution. The first component, which is a random algorithm, is to declustering databases initially among multiple processing nodes. The second component, which is composed by a splitting algorithm, a merging algorithm and two heuristic scheduling algorithms, is to support the dynamic adjusting of the database distribution among multiple processing nodes during running time. Experimental and theoretic results show that DCMD can efficiently support parallel processing of dynamic databases.