An Efficient Mapping Strategy for Parallel Programming

Obtaining an effective execution of a parallel system requires that the mapping of the processes (of the parallel software) on the processors (of the parallel hardware) is efficiently performed. Hence, this paper presents an efficient mapping strategy based on optimizing communications between processes as well as load balancing process distribution onto an arbitrary processor network. Such a mapping strategy is developed as a parallel program, based on the simultaneous execution of local, independent processes. This fact contrasts with many other approaches for solving the mapping problem, like simulated annealing, heuristic search, and others, which require a centralized control for the mapping. In this paper, it is shown that the present mapping strategy is efficient enough when applied to two different mapping problems. Based upon an experimental setup, it is possible to review this mapping strategy following the related impact.