Pittsburgh Supercomputing Center
NEWS RELEASE

FOR IMMEDIATE RELEASE                         CONTACT:
November 27, 1996                                Michael Schneider
                                                 Pittsburgh Supercomputing Center
                                                 412-268-4960
                                                 schneider@psc.edu


Pittsburgh, Oak Ridge, Sandia Team Wins Concurrency Gold


HPC Challenge Honors Materials Science Metacomputing.

PITTSBURGH -- A collaborative effort among three of the nation's leading computational research centers won a Gold Medal in the Concurrency category of the High Performance Computing Challenge at Supercomputing '96. A team of physicists, software developers and networking experts from Oak Ridge National Labs, Sandia National Labs and the Pittsburgh Supercomputing Center linked disparate, highly parallel computing systems at their respective sites. They applied this "metacomputing" approach to a large-scale materials science computation. The HPC Challenge judges cited them for demonstrating the ability to solve significant problems using diverse systems linked over high-speed networks.

"We've been talking about metacomputing for many years," said Sergiu Sanielevici, PSC parallel applications manager, "but actually producing science using geographically distributed systems is very hard to do. This tells us we're on the right track."

The Oak Ridge, Pittsburgh, Sandia team used high speed ATM networks to link Intel Paragons at Oak Ridge and Sandia with PSC's CRAY T3D. Using a novel first-principles materials science method, the Locally Self-Consistent Multiple Scattering method (LSMS), developed at Oak Ridge, they modeled ferromagnetic properties of a copper-nickel alloy. LSMS efficiently simulates the complex atom-to-atom interactions in a metal alloy by assigning each atom to a separate processor. It has demonstrated realistic results with structures of several hundred atoms, but larger structures, composed of thousands of atoms, must be simulated to reproduce neutron-scattering experiments on metal alloys, requiring more computing power.

For their HPC challenge demonstration, the collaboration designed and implemented network hardware and software that allowed the Intel and Cray systems to communicate with each other, exchanging data at the required rate in real time. They applied this metacomputing system to a 2916-atom structure of copper-nickel.

"This crowns several months work of effective team effort," says Sanielevici, "but the real importance is that this is an ongoing project, producing scientific insights that can't be obtained using any single machine. We're developing software, infrastructure and expertise -- at the applications and systems level -- to enable production runs with linked systems. This will benefit the national research and Department of Energy communities, who increasingly face problems that can only be solved with metacomputing."

More information (and graphics) about this award-winning project is available on World Wide Web:

http://www.ccs.ornl.gov/SC96/SC96.intro.html

The goals of the High Performance Computing Challenge, an annual competition sponsored by the annual Supercomputing conference, held this year in Pittsburgh, Nov. 17-22, are to foster approaches to teraflop (a trillion floating-point operations per second) performance and effective use of heterogeneous computing.

The Pittsburgh Supercomputing Center, a joint effort of Carnegie Mellon University and the University of Pittsburgh together with Westinghouse Electric Corp., was established in 1986 by a grant from the National Science Foundation, with support from the Commonwealth of Pennsylvania.

# # #

go back to contents page