Variable Sized Partitions for Range Query Algorithms

Abstract

A range query applies an aggregation operation over all selected cells of an OLAP data cube where selection is specified by the range of contiguous values for each dimension. Many works have focused on efficiently computing range sum or range max queries. Most of these algorithms use a uniformly partitioning scheme for the data cube. In this paper, we improve on query costs of some of these existing algorithms by noting two key areas. First, end-user range queries usually involve repetitive query patterns, which provide a variable sized partitioning scheme that can be used to partition the data cubes. Query costs are reduced because pre-computation is retrieved for entire partitions, rather than computed for a partial region in many partitions, which requires large amounts of cell accesses to the data cube. Second, data in the data cube can be arranged such that each partition is stored in as few physical storage blocks as possible, thus reducing the I/O costs for answering range queries.