Implementing MM5 on NASA Goddard Space Flight Center computing systems: a performance study

Abstract

We analyze and test the performance of the fifth-generation PSU/NCAR mesoscale model MM5 on parallel computers at NASA Goddard Space Flight Center. We show how MM5 code scales on the Cray J90, the Cray T3E and a cluster of PCs. More precisely, we are interested in finding the elapsed time, load balancing, speedup, number of floating point operations per second, and performance versus cost. Results obtained with two test problems show the efficiency of MM5 on the above computers especially with large size problems.