Early Experiences with the Cray T3E at the SDSC

Joel E. Tohline and John Cazes

Department of Physics & Astronomy
Louisiana State University

[e-mail: tohline@rouge.phys.lsu.edu or cazes@rouge.phys.lsu.edu]

Early Experiences with the Cray T3E
at the San Diego Supercomputer Center

In response to a request we received from Jay Boisseau (boisseau@SDSC.EDU) in April, 1997, we provide here a broad overview of our early experiences porting a computational fluid dynamics code to the Cray T3E at the San Diego Supercomputer Center (SDSC).

Part I: Background [May 8, 1997]
Type of Code ported.
Parallel Platforms on which our code has been executed previously.
Timings on several different machines.
Part II: Scalability Using the Portland Group HPF Compiler (The Good News) [May 8, 1997]
Parallel Programming Method.
Porting to the T3E.
  • Necessary code modifications
  • Problems we have faced
Performance Measures.
Part III: Performance of DEC Alpha processors (The Bad News) [June 6, 1997]
The Bad News.
Single-Processor Test Results.
Part IV: Summary Points [June 22, 1997]

Part IV: Summary Points
Successes, to Date:

  • The PGHPF compiler produces executable code that scales extremely well on the T3E, all the way from 2 to 128 processor nodes.

  • Using the T3E in conjunction with the PGHPF compiler, it now is possible to routinely perform CFD simulations at lattice resolutions of 2563 via a fortran90 code that contains no explicit message-passing statements and that can be ported fairly easily to a variety of other hardware platforms.

Room for Improvement:

  • The PGHPF compiler (which utilizes the F90 compiler on the T3E) produces executable code that performs computational instructions on each T3E processor with extraordinarily low efficiency. We suspect that this is not due to any major failings of the PGHPF compiler but, instead, can be ascribed primarily to the immaturity of the F90 compiler that currently is available on the T3E.

  • Based on our experience with Digital's F90 compiler for the 275 MHz Alpha processor, in the near future we should expect to see improvements in the F90 compiler on the T3E that lead to at least a factor of 3 improvement in execution speeds without having to make any major fortran code modifications.

  • Better tools are needed on the T3E to assist the user in optimizing HPF programs.

This work is supported in part through an NSF-funded project entitled, "Star Formation in Galaxies."