Estimating parallel performance, a skeleton-based approach

Abstract

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 without 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 program 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.

Three Dimensional Reconstruction

“…We have only stained endothelial cells and thus we cannot unequivocally differentiate very small arterioles (metarterioles) from capillaries in the sequence of the 21 serial sections by morphological 2, HSV conversion of next serial section (3), interpolation between both serial sections: in this case the fourth of seven images is shown (4), final result of all filtering operations of interpolated images (5), areas different between interpolated images and result of filtering coloured red (6). Scale bar = 50 µm.…”

“…The remaining 13 videos are found in the repository (doi: 10.5281/zenodo.128861). For each small region, the following processing steps are visualised from left to right in the upper and lower part of the panel: original registered data (1), HSV data conversion (2), HSV data conversion of next serial section (3), interpolations between both serial sections resulting in 7 images (4), final result of all volume filtering operations of interpolated images (5), areas differing between interpolated images and the result of filtering visualised in red colour (6). Video is available under doi: 10.1371/journal.pone.0168173.s021.…”

Farms, pipes, streams and reforestation: reasoning about structured parallel processes using types and hylomorphisms

“…pthreads, OpenMP, etc. Algorithmic skeletons also provide good structural cost models [14,20]. In this paper, we will use four basic parallel skeletons, each of which operates over a stream of input values, producing a stream of results: task farms, pipelines, feedbacks, and divide-andconquer.…”