The Brown University News Bureau
38 Brown Street / Box R
Providence, RI 02912
401 / 863-2476
Fax: 401 / 863-9595
1995-1996 index

Distributed July 17, 1995
Contact: Paula Korn

Brown takes a giant leap forward in computational science

PROVIDENCE, R.I. -- Researchers at Brown University's Center for Fluid Mechanics, Turbulence and Computation are embarking on a leading-edge multidisciplinary initiative in computational science with the acquisition of an IBM SP2 parallel supercomputer. Brown is one of a few sites in the United States with a computer of this magnitude, outside of national supercomputer centers.

IBM made possible the purchase of the $1.7-million supercomputer with a gift of $640,000, as part of the University's development campaign. The National Science Foundation also provided a $300,000 grant matched against University funding. The SP2 is a 24-node processor intended primarily for general and scientific computing. Scientific computing involves the simulation of complex physical systems, eliminating the need for costly laboratory experiments and allowing scientists to observe phenomena that otherwise are immeasurable or too costly. A multidisciplinary group of scientists at Brown is using the system to focus on four critical areas of research: parallel algorithms, turbulence simulations, computational vision and computational mechanics for materials design.

"The acquisition of the SP2, in addition to our existing computational facilities, makes the Center for Fluid Mechanics, Turbulence and Computation one of the strongest computational centers in the country, said George Karniadakis, principal investigator of the project and associate professor of applied mathematics. "These resources will allow researchers to address realistic simulations of turbulence in fluids." The system is based on IBM's UNIX implementation, AIX and RISC System/6000 technology which extends workstation economies to large system computing.

Co-investigators include Martin Maxey, director of the Center for Fluid Mechanics, and Donald McClure, chairman of the Division of Applied Mathematics. John Savage and Franco Preparata, professors of computer science, Michael Ortiz, professor of engineering, and David Gottlieb, professor of applied mathematics are also co-investigators. The system is maintained by Samuel Fulcomer, associate director for computer facilities and science visualization at the Center for Fluid Mechanics, who also is responsible for computer operations and network development.

The group is using the SP2 to study parallel computing for engineering and scientific applications. Because programming for parallel machines is challenging, such research requires techniques that exploit parallelism. Computer and computational scientists will work shoulder-to-shoulder in the SP2's multi-user operation mode to study parallel algorithm development, production computing and research training. Parallel processing enables simultaneous computation of many information points, or elements, of the same problem with speed and accuracy. The 24-node capability of the SP2 involves the linkage and interaction of 24 central processing units that may be programmed to work together on a specific application.

Turbulence simulation studies started with the birth of vector supercomputing (e.g. CDC, CRAY) in the late 1970s. Karniadakis said that with the emergence of parallel computing and soon-to-be-realized teraflop (one trillion floating operations per second) speeds, turbulence studies in a "numerical laboratory" will become part of the design phase. Unlike other "grand challenge" problems in physics, progress on the basic theory of turbulence translates almost immediately into a wide range of engineering applications and technological advances that affect many aspects of everyday life, such as environmental pollution, global change, fuel consumption and energy consumption.

"We anticipate that having a state-of-the-art parallel computer here at Brown," Karniadakis said, "will foster very close collaborations among departments and diverse research groups and will provide the means for introducing new courses in scientific computing. We have already pursued such an effort working closely with the Department of Computer Science."

The principal investigators and their respective research groups are combining expertise in computer sciences, numerical analysis, mathematical physics and computational mechanics. One of their primary objectives is to develop new parallel software based on algorithms, or computing processes, custom designed for efficient parallel simulations of turbulent flows at high speed, three-dimensional computational vision and image processing, and three-dimensional microstructures for materials design. While some of the applications may require even greater computational resources, which are available at the nation's supercomputer centers, a major part of smaller-scale production runs can be accomplished at Brown in a dedicated run mode, saving significant time and effort.

Experience has demonstrated that when new supercomputer architecture is introduced, it takes nearly a decade for mathematical software to mature on the new architecture. Slow development does not allow full exploitation of new computer architectures for engineering and science applications. Consequently, scientists make the greatest progress in the short term by tailoring advanced scientific libraries and virtual parallel environments for specific applications with the integration of low-level and high-level software.

Research with this computer is expected to generate benefits beyond immediate computational science. It will create an infrastructure for the development and exchange of algorithms from various fields of scientific and engineering computing. It also will engage other research groups in re-examining the efficiency of their algorithms and tackling new problems that have been computationally prohibitive in the past. Other Brown faculty in such diverse disciplines as chemistry, statistics, computer graphics, physics and engineering also have access to the SP2 through the Brown University Center for Scientific Computation, a center organized to support scientific computation. In addition to faculty, about 20 post-doctoral and 30 graduate students are using the parallel computer for completion of their Ph.D. programs.

Brown established the Center for Fluid Mechanics, Turbulence and Computation in 1986 to promote research in fluid mechanics, the study of turbulent flows and transition phenomena, the use of scientific computation in fluid dynamics research and topics of related scientific interest. Karniadakis joined the Brown faculty in January 1994 and is chairman of the NSF Peer Review Board for supercomputing allocation, associate editor of the Journal of Fluids Engineering and a member of the editorial board of the CFD Journal. His research interests focus on computational fluid dynamics with emphasis on turbulence simulations, high-order numerical methods, parallel processing and micromechanics and microfabrication. Additional information is available from the Center for Fluid Mechanics.