Delcstering Schemes with Guaranteed Performance

Abstract

Declustering schemes reduce response time of multidimensional range queries by distributing the data blocks across multiple disks. Previous declustering methods are mostly heuristic -based and provide neither guarantee nor asymptotic analysis of the performance. In this paper we seek declustering schemes with guaranteed worse deviation from the optimal. We devise an O (M4)algorithm, where M is the number of disks, to find the optimal (in worse case sense) scheme in the class of recently proposed cyclic declustering schemes. We also propose a new scheme based in O (M log M). We show by analysis that, for 2-dimensional data, the new scheme has a worst case response time within a factor 3 of the optimal for any query and within a factor 1.5 of the optimal for large queries. Furthermore, we show that the actual performance of the new scheme is even better than the theoretical bounds: it is at most 1 more than the optimal for M ≦ 22, and is at most 4more than the optimal for M ≦ 550.