Benchmarking in the Computer Systems Lab
Spring 2004
- Top500 List,
Top500.org (main page)
Press
release summary
Highlights
of the 22nd TOP500 List
HPCchallenge
Benchmarks (Jack Dongarra)
APEX
Application Performance Characterization and Benchmarking (Erich Strohmaier)
- Benchmark tests - a listing
Also - Benchmarking at NASA
AMES (Sheila Faulkner)
Resource
links from NASA Ames
- 1000s -
The LINPACK 1000x1000 benchmark program in single precision.
This works on our workstations, and the SV1.
Haven't tried it on the cluster yet.
Some results: Cray SV1,
Element,
Compaq,
Transformer
- HPL
A Portable Implementation of the High Performance Linpack Benchmark for
Distributed-Memory Computers.
This program can be used for the Top500 tests.
(Current status: I tried for several weeks to get this to compile on the SV1. It
finally compiled, but then had a runtime error.
I was coordinating with a researcher at U. Tenn, Knoxville)
- NAS -
A benchmark test program has been developed for use by the NAS program
at NASA Ames Research Center to aid in the evaluation of supercomputer performance.
NAS
documentation guide (pdf)
See - Cray SV1
benchmarking guide (p. 24) for tips on NAS code optimizing
Current status: This one works on the SV1 and our workstations.
(Haven't tried it on the cluster yet)
- Two from NASA Ames:
- "Hello World",
MPI version (works on the Cray SV1, haven't tried the cluster yet)
- MPI-TEST.tar, MPI
Benchmarking
Current status - have not successfully compiled for SV1
(haven't tried this on the cluster yet)
- Vector Architectures and Optimization for the Cray SV1
- Cray SV1:
- Benchmarking
guide for the SV1 (pdf)
- Hardware overview (p. 1 - 8): the processor, cache, peak and
measured bandwidths
- Vectorization (p. 8 - 19): dependencies, memory, stride, loop
unrolling,
choosing a loop ordering, fast intrinsics
- Parallel programming (p. 19 - 24): Autotasking/OpenMP, message passing
- MPI,
- NAS benchmark (p. 24 - 35): MXM - Matrix multiply, CFFT2D - Complex
fast fourier transform,
CHOLSKY - Cholsky decomposition, BTRIX - block tri-diagonal solver,
GMTRY - Gaussian elimination, EMIT - intrinsic functions
VPENTA - inverts 3 pentadiagonals simultaneously
- Summary of results for NAS (p. 35)
- Performance tools (p. 35 - ): hpm - hardware performance monitor,
profview - code profiling,
flowview - flow tracing, perfview - flow tracing with performance
counters,
fmaker - Make file generator, flipper - perl script,