Divide-and-conquer mapping of parallel programs onto hypercube computers

Abstract

Mapping of parallel programs onto parallel computers for efficient execution is a fundamental problem of great significance in parallel processing. This paper describes a heuristic algorithm for mapping arbitrary parallel programs onto hypercube computers using a divide-and-conquer technique. The running time of our algorithm is O(dn3), where n is the number of tasks in the parallel program and d is the dimension of the hypercube computer. The algorithm is implemented in C + + and its performance is evaluated through extensive testing and analysis.