Optimal parallel I/O for range queries through replication

Abstract

In this paper we study the problem of declustering two dimensional datasets with replication over parallel devices to improve range query performance. The related problem of declustering without replication has been well studied. It has been established that strictly optimal declustering schemes do not exist if data is not replicated. In addition to the usual problem of identifying a good allocation, the replicated version of the problem needs to address the issue of identifying a good retrieval schedule for a given query. We address both problems in this paper. An efficient algorithm for finding a lowest cost retrieval schedule is developed. This algorithm works for any query, not just range queries. Two replicated placement schemes are presented - one that results in a strictly optimal allocation, and another that guarantees a retrieval cost that is either optimal or 1 more than the optimal for any range query.