View online
Abstract
In this Redpaper we make recommendations for running the HPL version of LINPACK on IBM® xSeries® Linux clusters. This product is available from the following site: http://www.netlib.org/benchmark/hpl/
We consider three different interconnect technologies:
- Simple 10/100 ethernet
- Gigabit ethernet
- Myrinet interface with GM library
HPL uses the Message Passing Interface (MPI) to run a single LINPACK calculation in parallel across multiple servers. In this document we describe steps that you can take to achieve best performance on the Linux clusters.
Table of contents
Compiling and link-editing
Input file
Running LINPACK under MPICH-GM
Running LINPACK under public domain MPICH
Tuning for network performance
Results