Exploiting non-inherent parallelism in an internal mergesort

Abstract

Sorting and searching large data sets are two of the most widely re­ searched topics in the field of Com­ puter Science. Many sorting algorithms have been discovered, each with its own advantages and disadvantages (see [3] and [4] for examples). We present an algorithm which effectively utilizes a shared-memory, multi-processor to merge two large sorted lists. This algorithm exploits more parallelism than that inherent in the typical mergesort al­ gorithm and gets excellent 3peedups. We report the results of sorting a vec­ tor of 50,000 initially randomly or­ dered integers on each of four ma­ chines: a Sequent Balance 21000, a Sequent Symmetry, an Encore Multimax, and an Alliant FX/8.