An Evaluation of Architectural Platforms for Parallel Navier-Stokes Computations

  • Authors:

  • D. N. Jayasimha;M. E. Hayder;S. K. Pillay

  • Affiliations:

  • Intel Corporation, Santa Clara, CA 95052, djayasim@mipos2.intel.com;ICASE, NASA Langley Research Center, Hampton, VA 23681, hayder@icase.edu;Scientific Engg. Computing Solutions Office, NASA Lewis Research Center, Cleveland, OH 44142, spillay@lerc.nasa.gov

  • Venue:
  • The Journal of Supercomputing
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

We study the computational, communication, and scalabilitycharacteristics of a computational fluid dynamics application, whichsolves the time-accurate flow field of a jet using the compressibleNavier-Stokes equations, on a variety of parallel architecturalplatforms. The platforms chosen for this study are a cluster ofworkstations (the LACE experimental testbed at NASA Lewis), ashared-memory multiprocessor (the CRAY Y-MP), and distributed-memorymultiprocessors with different topologies (the IBM SP and the CRAYT3D). We investigate the impact of various networks connecting thecluster of workstations on the performance of the application and theoverheads induced by popular message-passing libraries used forparallelization. The work also highlights the importance of matchingthe memory bandwidth to processor speed for good single processorperformance. By studying the performance of an application on a varietyof architectures, we are able to point out the strengths and weaknessesof each of the example computing platforms.