Estimating Parallel Performance, A Skeleton-Based Approach?

Abstract. In this paper we estimate parallel execution times, based on identifying separate “parts ” of the work done by parallel programs which are defined using algorithmic skeletons. Our runtime analysis works with-out any source code inspection. The time of parallel program execution is expressed in terms of the sequential work and the parallel penalty. We obtain these values for different problem sizes and numbers of processors and estimate them for unknown values in both dimensions. This allows us to predict parallel execution time for unknown inputs and non-available processor numbers. Another useful application of our formalism is a measure of parallel pro-gram quality. We analyse the values for parallel penalty both for growing input size and for increasing the number of processing elements. From the behaviour of these data, conclusions on parallel performance and scalability are drawn