Design of a Parallel Computational Fluid Dynamics Code on a Shared Memory Architecture
Harris L. Bergman, Jeffrey S. Vetter, Karsten Schwan, David N. Ku · Advances in Bioengineering · 1995
Abstract Computational fluid dynamics (CFD) requires an enormous amount of computer time. While CFD is an alternative to experimentation, the CPU time requirements for three dimensional, unsteady problems are prohibitive. Investigations in this laboratory have investigated pulsatile flow through the coronary bifurcation using a spectral element method (SEM) algorithm; run times were on the order of six weeks on an IBM RISC/6000. Use of Cray supercomputers offered little benefit. A Cray C90 is a few times faster than a high-performance workstation when executing non-optimized code. Furthermore, users often wait days or even weeks for their jobs to reach the front of the supercomputer queue. Parallel execution is one way to gain performance with minimal cost.