A new class of two stage parallel sorting schemes

Abstract

A two stage parallel sorting scheme is presented in which in the first stage the input file is divided into a number of subfiles and sorted in parallel using the conventional heap sort algorithm. The second stage then merges the sorted sub-files in parallel. It is shown that a given input file of size n can be sorted in 0(n) time using 0(log n) processors. The speed-up ratio, which is a measure of the effectiveness of parallel processing, with respect to the best sequential algorithm, is asymptotically proportional to log n, which is optimal in the number of processors used.