文档介绍：Redbooks Paper Colin Dumontier
Oliver Hess
Francois Thomas
Building an HPC Cluster with Linux,
pSeries, , and Various Open Source
Software
Linux/pSeries® clusters for HPC (High puting) is a new offering that is
attracting more and more clients. Often, they are already using a Linux/xSeries® cluster and
would like to benefit from the outstanding 64-bit floating-point performance and capability
featured by Power4-based servers. This offering allows them to reap these benefits and
retain Linux® as their operating system.
This IBM® Redpaper explains how to bring a collection of pSeries nodes and a
interconnect to a state that permits running a true HPC production workload in a
Linux-clustered environment. Already very popular with Linux/xSeries clusters, is the
high speed and low work that we use for distributed memory applications on
Linux/pSeries clusters.
© Copyright IBM Corp. 2004. All rights reserved. /redbooks 1
Introduction
Although IBM is quickly bringing the well-proven software previously available for only AIX®
users to this space, the current offering still lacks some pieces to build, program. and operate
a Linux/pSeries HPC cluster. We will explore how the use of open source software can bridge
this gap (and possibly even remain when the full IBM software environment is ready for use).
Compared to the well-known AIX HPC environment, another major difference is the use of the
interconnect technology as the high speed and low work for MPI
applications. This brings up many questions that we will try to answer.
In this document, we explain how to bring a collection of pSeries nodes and a
interconnect to a state where we can run a true HPC production workload in a Linux-clustered
environment. IBM does not plan to enable its latest high-working technology for
AIX/pSeries clusters for Linux/pSeries. Therefore, , which is already very popular with
Linux/xSeries clusters, is the high speed