CSE SYMPOSIUM KEYNOTE ADDRESS

SPEAKER: PROF. PAUL R. WOODWARD

Department of Astronomy
University of Minnesota, Minneapolis, MN

DATE: FRIDAY, April 18, 1997

TIME: 3:00 PM

PLACE: 335 Grainger Engineering Library

TITLE: Simulating Turbulent Fluid Flow on Clusters of Shared Memory Machines

ABSTRACT

Nearly four years ago my colleagues and I at the University of Minnesota collaborated with Silicon Graphics to build a prototype machine, a cluster of 16 powerful, 20-processor shared memory machines (SMPs) which we called the Challenge Array, to simulate homogeneous turbulence on a grid of over a billion cells with our PPM gas dynamics code. For this experiment, we restructured our code to exploit the shared memory inside each cluster member while passing messages between the SMPs as in a distributed memory machine. By now, this hardware configuration, the SMP cluster, has entered the supercomputing mainstream, especially at national centers like NCSA and Los Alamos. However, the programming techniques which should go along with this hardware are not yet in common practice. I will briefly review the familiar methods for constructing multi-tasked parallel programs for shared memory machines. Then I will suggest an approach to extending these techniques over a cluster of SMPs, whether the cluster has distributed shared memory (DSM) hardware support for such programs or not. Finally, I will show some examples of calculations that my research group has carried out on SMP clusters and indicate the still more exciting computations that we expect to be supported by the new generation of these machines, like those being acquired by NCSA.

BIOGRAPHY

Paul R. Woodward was born on August 25, 1946, in Rockeville Centre, New York. He received a B.A. in Mathematics and Physics from Cornell University in 1967 and a Ph.D. in Physics from the University of California, Berkeley, in 1973. He worked for many years at the Lawrence Livermore National Laboratory before taking, in 1985, his present position as Professor of Astronomy at the University of Minnesota. Professor Woodward is a Fellow of the Minnesota Supercomputer Institute and is Director of the University of Minnesota's Laboratory for Computational Science and Engineering. Professor Woodward's research has focussed on the exploration of nonlinear phenomena in fluid dynamics through large-scale computer simulations. He has worked principally on turbulent flows, and especially in astrophysical contexts where the effects of compressi- bility are important. A persistent theme in his work is the use of computers to perform numerical experiments for fluid dynamical systems where the governing laws of physics are understood but their implications are not, due to the complexity of chaotic nonlinear interactions which develop. To enhance the predictive power of these experiments, Woodward has worked on the development of accurate numerical techniques, the best known of which is the Piecewise-Parabolic Method (PPM) for compressible gas dynamics. To make the results of the experiments more easily understandable, Woodward has led the development of powerful scientific visualization systems and software. His efforts in scientific visualization have centered on the special problems presented by the extremely large data sets which his numerical experiments have produced.