Performance Prediction of Parallel - Methodology Systems and Haring with Scalable Specifications Case Study* Harald Wabnag, University GGnter of Vienna Institute of Applied Lenaugasse Computer 2/8, .univie. Science A-108O, ac. at, and Vienna, Information Austria .univie. ac. at Systems wabnig@ani haring~ani In parallel neering gram. load cessor not software development to get the performance parallel algorithm the engiproor the load nodes are mapped to different are the processors. execution graph task Worktimes nodes of and arcs. hard- is important If, for example, to an efficient parallel match resulting description parameters code represented choosen does not the the sequential the amount by task across balancing system the to strategy characteristics, full multiprowill It early of data parameters routing the that is sent the graph of the software system. in Resource ware, type, values The crete the and describe strategy, and the topology exploit processing power of the estimates communication software network parameter is crucial stages the sign determine parallel from performance program making hardware system of the development wrong parallel to prevent program de- related PAPS to communication. toolset timed Basic and automatically Petri Petri arcs net net out generates of the a disparallel are developers decisions. The PAPS
/lp/association-for-computing-machinery/performance-prediction-of-parallel-systems-with-scalable-x0WK5LwBJ2