Issue No. 01 - January (1991 vol. 2)
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/71.80193
<p>New methods are presented for bounding and approximating the mean execution time ofpartitioning algorithm, and these methods are compared to previous approaches.Distribution-driven and program-driven simulations show that two of the methods areusually accurate to within 10% and give good estimates even when certain independenceassumptions are violated. Asymptotic approximations and upper bounds are derived forthe average execution time of multiphase algorithms when there is no contention forprocesses in the parallel phase. In addition, the authors bound the average executiontime under static and dynamic scheduling policies and determine the optimum number ofparallel tasks to be created to minimize the execution time bounds with constantscheduling overhead.</p>
Index Termsdistribution driven simulations; asymptotic approximations; synchronous parallelalgorithms; regular structures; bounding; mean execution time; partitioning algorithm;program-driven simulations; upper bounds; average execution time; multiphasealgorithms; scheduling policies; parallel tasks; execution time bounds; parallel algorithms; parallel programming; performance evaluation; scheduling
S. Madala and J. Sinclair, "Performance of Synchronous Parallel Algorithms with Regular Structures," in IEEE Transactions on Parallel & Distributed Systems, vol. 2, no. , pp. 105-116, 1991.