Skip to main content

IBM Systems  >   System p  >   Hardware  >  
 

IBM eServer pSeries High Performance Switch

Performance white paper

 

In this paper, we describe the performance results for several communication benchmarks running on the HPS-based (IBM eServer™ pSeries® High Performance Switch and adapter) systems. We provide some guidelines for end users on the various environmental settings that can be tuned to improve the performance of applications. The specific measurements are for the MPI and TCP/IP communication protocols. The MPI measurements were done using a simple ping-pong based benchmark which we call Spark. For TCP/IP, we use an internal benchmark similar to the Netperf public-domain benchmark.

Introduction
In this paper, we first provide an overview of the HPS switch and adapters, pSeries server characteristics that impact performance of parallel applications using MPI and IP. Application characteristics and the time the server takes to process the communication protocol stack, determine how much of the HPS performance capability can be realized by parallel applications.

We also we provide a brief overview of the switch and associated building blocks of the overall high performance computing (HPC) cluster, and describe our own experiences in benchmarking MPI and TCP/IP performance on these systems. Several different types of pSeries servers were used in this study including the pSeries 655 with 1.7 GHz POWER4+™ processors and the pSeries 690 with either 1.7 GHz or 1.9 GHz POWER4+ processors. We also used the more recent ~ p5 575 with 1.5 and 1.9 GHz IBM POWER5™ processors. Measurements were done using both AIX 5L™ V5.2 and V5.3.

The rest of the paper is organized as follows. In Section 2, we describe the high level architecture and design of the HPS switch. In Section 3, we provide detailed performance results of our experiences running our MPI benchmark. Some general observations about environment settings and an outline of how users can make use of them are provided. TCP/IP performance results are provided in Section 4.

For the measurements in this paper, we used HPS POWER5 SP 3 for POWER5 processor-based servers and include historical results from the HPS POWER4™ SP 11 release of the HPS code for POWER4+ processor-based servers. There are separate results for each level of the HPS code. Not all hardware configurations were run at all Service Pack levels. Throughout this document, adapter refers to any adapter type, unless a specific adapter is named.

Full text white paper
View Acrobat
(104 KB) January 2006
 
Get Adobe® Reader®