A strictly-optimal strategy to access multi-dimensional data on parallel disk systems

The disk allocation problem addresses the issue of how to distribute large files among several disks so as to maximize the concurrent disk accesses in response to partial match queries. In the past, this problem has been studied for binary as well as for p-ary cartesian product files. We propose a strictly-optimal disk allocation strategy for non-uniform cartesian product files for every partial match query. Our strategy is based on a large and flexible class of maximum distance separable (MDS) codes, namely the redundant residue codes. A new family of residue codes, called the redundant non-pairwise prime residue codes, is also introduced