Scaling Performance of MM5 on the Cray T3E and a Beowulf-Like System

Jules Kouatchou
School of Engineering
Morgan State University
5200 Perring Parkway
Baltimore, MD 21239
and NASA

Abstract

We present the performance of the fifth-generation PSU/NCAR mesoscale model MM5 on distributed-memory parallel computers in particular the Cray T3E and a cluster of PCs at NASA Goddard Space Flight Center. We are interested in finding the elapsed time, load balancing, speedup, number of floating point operations per second as function of the number of processors. Results obtained with three test problems of different sizes show the efficiency of the MM5 code on both systems and also a noticeable scaling improvement as the problem size increases.



Return to High Performance Scientific Computations with Applications main page.