Determining Optimum Software Partitioning and Data Allocation on a Mimd System to Minimize the Completion Time

In a parallel environment, where processes reside on diierent processors and work on the same computation, data allocation plays a signiicant role in the completion time. Data allocation depends on the way software is partitioned; therefore, both the data and software should be considered together to achieve a global optimization. In this paper, we present an algorithm that produces the optimum software partitioning and data allocation for a parallel computation. The algorithm incorporates the retrieval and update behavior of data items, the allocation of data to processes, the execution times of processes, the control and data precedences, and the communication time of the target multiprocessor. While partitioning the software and allocating the data, the algorithm minimizes the interprocess communication time thus optimizing the overall completion time.