On deciding granularity for optimal speedup for solving data parallel problems with clustered distributed computing
Jung‐Sing Jwo, Yu Chin Cheng, Chin-Yun Hsieh · 2002
In this paper we show how to obtain optimal speedup in a master-slave model for solving data-parallel problems. Given the number of homogeneous workstations, their computation time for solving a basic sub-task of the problem, network transmission bandwidth and data volume per basic sub-task, the per-distribution number of basic sub-tasks sent to a slave for attaining the optimal speedup can be decided. The effectiveness of the proposed theory has been tested using a parallel computing experiment involving the Hough transformation.