Redbooks Paper
Effective System Performance Suite on
IBM Eserver pSeries
This Redpaper highlights the improvement in Effective System Performance achieved on the
IBM Eserver pSeries® 655 system. By executing the latest version of ESP suite and using
appropriate LoadLeveler® settings, superior performance ratios for ESP (94%) were
achieved.
The authors of this Redpaper
This Redpaper was produced by a team of specialists from around the world.
Sheeba Prakash is a staff Engineer/Scientist at the IBM Eserver pSeries and HPC
Benchmark Center. She joined the HPC benchmarking team in 2000 and works on
performance analysis of scientific and technical applications for pSeries and xSeries®
servers. She holds a master's degree in Traffic Engineering and Transportation Planning and
a Master of Science in Computer Science from the University of Houston.
Barry Spielberg is a certified AIX® System Administrator working in the pSeries Benchmark
Center. He manages large AIX clusters used for commercial and technical application
benchmarking.
Sheeba Prakash
Barry Spielberg
© Copyright IBM Corp. 2003. All rights reserved.
ibm.com/redbooks
1
Introduction
Effective System Performance (ESP) test is a National Energy Research Scientific
Computing Center (NERSC)-designed benchmark suite that provides a metric that focuses
on operating system attributes for production-oriented parallel systems. These attributes
include parallel launch time, job scheduling, preemptive job launch, and system reboot time.
The metric is processor speed independent with low contention for shared resources such as
the file system. The mechanics of the test is similar to that of a throughput measurement, but
is not a throughput test.
The overall value of a High Performance Computing system depends not only on the raw
computational speed, but also on the efficiency of the system management. LINPACK and
NAS parallel benchmarks are useful for measuring sustained computational performance, but
they give little or no insight into system-level effi