Top-down Heuristic for Finding Optimal Grain Size of Parallel Tasks

In order to have an optimal execution time of a program running on a multiprocessor system, the program has to be partitioned into concurrent tasks. Partitioning of programs to grain size suitable for parallel execution is an NP complete problem but near-optimal time can be derived. This paper discusses a heuristic to determine the near-optimal grain size of parallel tasks that will give the best execution time. The effects of communication overheads between the different processors are examined. The heuristic developed is capable of balancing between maximizing parallelism and minimizing overheads