Jarrett Cohen
Goddard Space Flight Center, Greenbelt, Md.
(301/286-2744, jarrett.cohen@gsfc.nasa.gov)

Microgravity Code Surpasses 50 GigaFLOPS on Goddard CRAY T3E

University of Texas at Austin researchers' microgravity code sustained 51.6 billion floating-point operations (gigaFLOPS) on the 512-processor CRAY T3E supercomputer at NASA/Goddard Space Flight Center in Greenbelt, Md. A subsequent run on a 1,024-processor CRAY T3E-900 (the processors are 50 percent faster) produced a rate of 112.8 gigaFLOPS. The code enables high-resolution supercomputer simulations aimed at improving understanding of fluid flows in microgravity. These simulations will be used to test manufacturing in space and critical functions of the International Space Station.

<MGFLO> is a three-dimensional coupled heat transfer and viscous flow application employing the finite element approach. "Our studies are designed to probe the fundamental nature of fluid flow in a microgravity environment, where thermocapillary surface tension effects are dominant," said team leader Graham Carey, professor of aerospace engineering and engineering mechanics at UT. In the near absence of gravity, temperature changes pull the fluid through surface tension.

"Recent terrestrial experiments on thin films by our collaborator Harry Swinney and his associates have described some interesting new nonlinear interactions associated with the free surface effects," Carey said. "Those interactions will be particularly important in a microgravity environment, but little is known about them at this point. Our high-performance computations will permit accurate, reliable modeling at the resolution necessary to capture the fine-scale nonlinear interactions."

The version of <MGFLO> is essentially the same as that achieving 16.5 gigaFLOPS on the 512-processor CRAY T3D, resulting in greater than three-fold improvement on the CRAY T3E. In addition to raw performance, "scalable parallel efficiency is an important component to fine-grid, high-resolution, three-dimensional simulations," Carey said. His team's application scales essentially linearly from one to 512 processors, with over 95 percent scaled speed-up.

Sustained parallel performance above 50 gigaFLOPS is the second performance milestone in UT's and eight other Grand Challenge cooperative agreements with NASA's High Performance Computing and Communications Program. By 1999, the Earth and Space Sciences Project Grand Challenge teams must reach 100 gigaFLOPS sustained on their codes.

The code is in its first phase of development, having begun in summer 1996. Science studies with the current and future versions of the code will help ascertain the effectiveness of manufacturing higher quality thin films and coating processes as well as growing crystals in space. The code is also relevant to the functioning of the Space Station's life support and safety systems.

The computational team for the benchmark effort were Carey, Christopher Harle and Robert McLay of the aerospace engineering and engineering mechanics department. Related parallel studies are being carried out by Carey with Bruce Davis at Goddard and Abhijit Bose at UT. Spencer Swift is the Cray Research, Inc. support engineer for the project at Goddard, where the benchmark studies were verified. Swinney and Stephen van Hook at UT's Center for Nonlinear Dynamics are performing related experiments on thin liquid layers.

More information about this research is available at the following URL:

http://www.cfdlab.ae.utexas.edu/otherresearch/nasa_hpcc/