K.K. Yue, Dept. of Comput. Sci., Minnesota Univ., Minneapolis, MN, USA
D.J. Lilja, Dept. of Comput. Sci., Minnesota Univ., Minneapolis, MN, USA
Proposes a novel methodology to develop new parallel loop scheduling algorithms. First, existing algorithms are parameterized into a general framework. Then, a genetic algorithm is used with a shared-memory multiprocessor simulator to "evolve" better strategies. Two new algorithms (called CS-2 and FS-alt, due to their similarity to chunk scheduling and factoring, respectively) are found that perform well compared to existing algorithms.
Index Terms:
parallel algorithms; parameter estimation; scheduling; genetic algorithms; virtual machines; shared memory systems; program control structures; parameter estimation; generalized parallel loop scheduling algorithm; genetic algorithm; multiprocessor simulator; strategy evolution; CS-2; FS-alt; chunk scheduling; factoring; shared-memory multiprocessor
Citation:
K.K. Yue, D.J. Lilja, "Parameter estimation for a generalized parallel loop scheduling algorithm," hicss, pp.187, 28th Hawaii International Conference on System Sciences (HICSS'95), 1995