Parallelizing divide-and-conquer algorithms — Microtasking versus autotasking

Abstract

Algorithms based on a divide-and-conquer strategy are well qualified for being implemented in a multitasking environment. The idea of the divide-and-conquer paradigm is to fragment a problem into subproblems of the same kind, to solve the subproblems recursively, and, finally, to combine the solutions of the subproblems into a solution of the original problem. The subdivision in smaller problems which can be solved independently provides the possibility for parallel execution on multiple processors. In this paper the parallel implementation of a divide-and-conquer algorithm to compute the convex hull in the plane is discussed. The algorithm is implemented in FORTRAN on a CRAY Y-MP8/832 using the CRAY multitasking strategies. The concepts of microtasking and autotasking are compared with respect to their qualification for the parallelization of REPEAT loop constructs which constitute the main part of the described divide-and-conquer algorithm.