MPIBench, version 1.1

What is MPIBench:

MPIBench is a tool for detailed measurement of the performance of MPI communications routines.

Features:

MPIBench has several features that distinguish it from other MPI benchmarks, such as the Pallas MPI Benchmarks, the PARKBENCH/Genesis low level communication kernels, MPBench, Mpptest, SKaMPI and others.

In particular it:

• uses a highly accurate global clock for measurement.
• times individual routine calls, at all processes.
• generates timing distributions of individual calls.

MPIBench also:

• generates summary output.
• can be easily modified to measure more communications calls and patterns.

Requirements:

The timing code in MPIBench requires:

• access to a (local) clock register (most modern CPUs support this).
• kernel support to bind a process to a physical processor if SMP nodes are used (since time stamp counters are not usually synchronised).

Installation and Use:

• MPIBench source code

• MPIBench User Guide

Publications:

• Comparison of MPI Benchmark Programs on an SGI Altix ccNUMA Shared Memory Machine, Nor Asilah Wati Abdul Hamid, Paul Coddington and Francis Vaughan, Proc. of Workshop on Performance Modeling, Evaluation, and Optimization of Parallel and Distributed Systems (PMEO-PDS'06), Rhodes, Greece, April 2006. DHPC-169
• Performance Analysis of MPI Communications on the SGI Altix 3700 , Nor Asilah Wati Abdul Hamid, Paul Coddington and Francis Vaughan, Proc. of APAC'05, Gold Coast, Sep 2005. DHPC-160
• Communication Benchmarking and Performance Modelling of MPI Programs on Cluster Computers , D.A. Grove and P.D. Coddington, International Journal of Supercomputing, vol. 34, 201-217 (2005). DHPC-143
• Analytical Models of Probability Distributions for MPI Point-to-Point Communication Times on Distributed Memory Parallel Computers , D.A. Grove and P.D. Coddington, Proc. of 6th International Conference on Algorithms and Architectures for Parallel Processing (ICA3PP-2005), Melbourne, Oct 2005, Lecture Notes in Computer Science, Volume 3719, pp 406-415, Springer, 2005. DHPC-163
• Performance Modelling of Message-Passing Parallel Programs , Duncan Grove, PhD Thesis. DHPC-138
• Performance Analysis of MPI Communications on the AlphaServer SC , D.A. Grove and P.D. Coddington, Proc. of APAC'03, Gold Coast, September 2003. DHPC-139
• Communication Performance Issues for Two Cluster Computers , Francis A. Vaughan, Duncan A. Grove and Paul D. Coddington, Proc. of 26th Australasian Computer Science Conference (ACSC 2003), Adelaide, February 2003. DHPC-137
• Precise MPI Performance Measurement Using MPIBench , D.A. Grove and P.D. Coddington, Proc. of HPC Asia 2001, Gold Coast, September 2001. DHPC-104