Latin Hypercubes: A Class of Multidimensional Declustering Techniques

The I/O subsystem is widely accepted as one of the principal bottlenecks for high performance parallel databases systems. The emergence of parallel I/O architectures has made the problem of data declustering, i.e. fragmenting a le of records and allocating the pieces to different disks, one of prime importance. This is evident from the growing activity in this area. In this study we focus only on multi-attribute declustering methods which are based on some type of grid-based partitioning of the data space. Since the multidimensional range query is the main workhorse for applications accessing such data, the focus is to provide eecient support for it. We rst show that there exists no declustering method that is strictly optimal for range queries if the number of disks is greater than 5. Thus the focus is on using declustering methods which provides good average case performance and are also optimal for a large class of queries. A class of multidimensional declustering methods, called Latin Hypercubes, is proposed. Conditions under which this class is optimal are derived. Also provided are the worst case and average case bounds on multidimensional range query performance. A detailed experimental evaluation is carried out to see how the class compares with other declustering methods. Parameters that are varied are shape and size of queries, database size, number of attributes and the number of disks. Our ndings (theoretical and experimental) show that latin hypercubes do very well for large queries (near optimal), and partial match queries, and are within reasonable bounds of other declustering methods for small queries. Since it is not possible to have a declustering method which performs optimally for all possible range queries, our ndings help decide when to use this class of methods. Finally, since there is no clear winner, parallel database systems must support a number of declustering methods and Latin Hypercubes would invariably have to be one of them.