Microtasking general purpose partial differential equation software on the CRAY X-MP

Abstract

Microtasking on four-processor CRAY X-MP computers is examined. The examination includes descriptions of important properties of Cray microtasking and an explanation of how to apply it to a program package. The experience of applying Cray microtasking to the general purpose partial differential equation package PLTMG of R. Bank et al. [1985, 1986] is communicated. Results of experiments in a single-job (dedicated) computing environment show that a large part ofPLTMG's interprocedural parallelism can be exploited with microtasking and little to moderate human effort. An experiment involving PLTMG and conducted to assess the performance of Cray microtasking in a multiprogramming (batch) environment is also described. This and related experiments described in [Bieterman 1987] are evidently the first of their kind reported.

Multitasking a Navier-Stokes algorithm on the CRAY-2

“…Results of microtasking were first reported by Misegades et al [1987], who reported speedups of 3.40 and 3.65 on the CRAY X-MP/4 for two CFD codes. Bieterman [1987] reported speedups of 1.22 to 2.36 on the CRAY X-MP/4 using microtasking for the partial differential equation package PLTMG. On the CRAY-2, Swisshelm [1988] reported a speedup of 2.67 using macrotasking for a three-dimensional Navier-Stokes algorithm.…”