Scheduling of multiprocessor tasks for numerical applications

Many applications in the area of scientiic computing use algorithms with a two-level parallelism based on potential method parallelism and potential system parallelism. In this paper, we investigate the eecient implementation of such algorithms on distributed memory machines. We consider parallel speciications consisting of an upper level of multi-processor tasks each of which having an internal structure of uni-processor tasks. To achieve an optimal parallel execution time, the parallel execution of such a program requires an optimal scheduling of the multi-processor tasks and an appropriate treatment of uni-processor tasks. In particular, we consider an important class of parallel programs that are generated within a speciic parallel programming model designing group-SPMD programs for scientiic computing. We show how the costs of data redistributions between M-tasks can be taken into consideration and how the special structure of the resulting program can be exploited by using a simple approximation algorithm with a provable good performance.