Parallel sorting on a shared-nothing architecture using probabilistic splitting

Abstract

The authors consider the problem of external sorting in a shared-nothing multiprocessor. A critical step in the algorithms the authors consider is to determine the range of sort keys to be handled by each processor. They consider two techniques for determining these ranges of sort keys: exact splitting, using a parallel version of the algorithm proposed by Iyer, Ricard, and Varman; and probabilistic splitting, which uses sampling to estimate quantiles. They present analytic results showing that probabilistic splitting performs better than exact splitting. Finally, the authors present experimental results from an implementation of sorting probabilistic splitting in the Gamma parallel database machine. >