Analysis of execution and communication times of multiprocessor tasks

Much research has been done into the allocation of precedence-constrained tasks to a parallel or distributed processor computer. An accurate a priori measure of each task's execution time and the inter-task communication times is necessary in order to find an optimal or near-optimal solution; however, many previous papers have used estimates of the execution and communication times, presenting the likelihood of sub-optimal task-to-processor allocations. In this paper, we present a multiprocessor programming environment which has been used successfully to model algorithms as directed acyclic task precedence graphs, and to generate multiprocessor implementations of the algorithms which can accurately determine task execution times and inter-task communication times. Once these times are determined, a near-optimal allocation can be determined easily with well-established branch-and-bound scheduling methods.<
>