Perfect Allocation Methods for Spatial Queries in Parallel Disk Systems

A disk-allocation method assigns a disk-id to each unit of spatial data. Allocating spatial data over multiple disks to distribute the I/O cost of query processing uniformly over available disks can tremendously speed up the processing. An allocation method is called perfect for a query set if it balances the I/O load on each disk in processing any query in a query set. Some of the interesting query sets in spatial databases include path queries and orthogonal range queries. We provide a perfect allocation method for all path and orthogonal range queries for five and fewer disks. A non-intuitive result in this paper shows that there is no perfect allocation method for the set of orthogonal range queries for more than five disks. We examine the effectiveness of Latin Square and Linear allocation methods for the interesting query sets in this context. We show that Latin Square allocation methods can be perfect for a class of orthogonal range queries and linear allocation method can be perfect on a class of path queries as well as many classes of orthogonal range queries.