Replicated Parallel I/O without Additional Scheduling Costs

Abstract

Abstract.A common technique for improving performance in a database is to decluster the database among multiple disks so that data retrieval can be parallelized. In this paper we focus on answering range queries in a multidimensional database (such as a GIS), where each of its dimensions is divided uniformly to obtain tiles which are placed on different disks; there has been a significant amount of research for this problem (a subset of which is [1,2,3,4,5,6,7,8,9,11,12,13,14,15]). A declustering scheme would be optimal if any range query could be answered by doing no more than # of tiles inside the range/# of disks retrievals from any one disk. However, it was shown in [1] that this is not achievable in many cases even for two dimensions, and therefore much of the research in this area has focused on developing schemes that performed close to optimal. Recently, the idea of using replication (i.e. placing records on more than one disk) to increase performance has been introduced. [7,12,13,15]. If replication is used, a retrieval schedule (i.e. which disk to retrieve each tile from) must be computed whenever a query is being processed. In this paper we introduce a class of replicated schemes where the retrieval schedule can be computed in time O(# of tiles inside the query's range), which is asymptotically equivalent to query retrieval for the non-replicated case. Furthermore, this class of schemes has a strong performance advantage over non-replicated schemes, and several schemes are introduced that are either optimal or are optimal plus a constant additive factor. Also presented in this paper is a strictly optimal scheme for any number of colors that requires the lowest known level of replication of any such scheme.

Analysis and Comparison of Replicated Declustering Schemes

“…The only exceptions are partitioned allocation and dependent allocation. Partitioned allocation [18], [8], [2] divides the disks into pairs and replicates a disk on the other disk in the pair. Therefore, there are n 2 distinct possibilities for the pair of disks a bucket can be stored at.…”

“…This is a 3 Â 2 query with an optimal retrieval cost of d 3Â2 7 e ¼ 1. However, since in the first copy, the buckets [0, 0] and [2,1] are both stored on disk 0, retrieval using the first copy requires two disk accesses. When we consider both copies, we can represent the problem using a bipartite graph.…”

“…In partitioned replication [8], [17], [18], [2], the set of disks is divided into groups and disks in a group are replicated on other disks in that group. Each group has c members for a ccopy replicated declustering scheme based on partitioned replication.…”