site map Top Resources Other Sites Implementations Tools & Libraries Instructional MPI2 What's New Feedback

MPI Implementation Page

The University of Notre Dame Laboratory for Scientific Computing maintains a list of all publicly available MPI implementations including what features of MPI2 are supported, as well as vital statistics about these implementations.

Freely Available MPI Implementations

There are several freely available MPI implementations:
* MPICH
The Argonne National Lab / Mississippi State University implementation.
* MPICH-PM/CLUMP
The MPICH-PM/CLUMP MPI library for Myrinet supports clusters running Linux, NetBSD, and SunOS. Pentium SMP clusters are supported also.
* MPICH-T3E
MPICH-T3E is MPICH-1.1.0 ported to the Cray T3E by the HPC Lab at MSU. mpirun supports NQE.
*LAM
The LAM MPI implementation is now supported and maintained by the University of Notre Dame.
*Chimp
The Chimp implemenation of MPI was developed at the Edinburgh Parallel Computing Centre.
*Unify
The Unify subset implemenation of MPI was developed at Mississippi State University. It runs on top of PVM and provides the programmer with a dual-API; single programs can contain both MPI and PVM code.
*BDM/MPI
The BDM/MPI was developed at Mississippi State University. It uses reliable protocols over Myrinet for MPI communications.
*MPICH/NT
MPICH/NT is a full implementation of MPI for Windows NT workstations. It is based on MPICH and is available from Mississippi State University. This version supports both shared memory communications within a workstation and TCP/IP communications between processes running on multiple workstations concurrently. This is no longer actively supported.
*W32MPI
W32MPI is a full implementation of MPI for MS-Win32 Clusters It is based on MPICH and is available from the Instituto Superior de Engenharia de Coimbra - Portugal and the Universidade de Coimbra - Departamento de Engenharia Informatica - Portugal.
*WinMPI
WinMPI is the first MPI implementation for MS-Windows 3.1. It runs on a single IBM compatible PC, no networking is required. WinMPI is based on MPICH. It is provided by the University of Nebraska at Omaha.
*MPI-FM
MPI-FM is a high-performance port of MPICH for SPARCstation clusters interconnected by Myrinet. It is based on Fast Messages and comes from the University of Illinois at Urbana-Champaign.

Information on Other MPI Implementations

* MPI/Pro
MPI Software Technology, Inc. provides commercial MPI implemenations for clusters, Windows NT, and multicomputers.
* PaTENT MPI 4.0
Genias provides a commercial packaging of WMPI (Intel x86 Win32).
* ScaMPI
ScaMPI is a high bandwidth, low latency MPI implementation for UltraSPARC and x86 systems running Solaris 2.6. A Linux and Windows NT version will be available by June 1998.
* Pallas
Pallas is a leading independent software company specializing in High Performance Computing (HPC). Pallas assisted many organizations in migrating from sequential to parallel computing. Customers of Pallas come from all fields: hardware manufacturers, software vendors, end users. Each of these has benefited from Pallas's unique experience in the development and tuning of parallel applications.
In the field of MPI development tools and implementations, Pallas contributions comprise: VAMPIR is currently the most succesful MPI tool product (see also "Supercomputer European Watch", July 1997), or check references at http://www.cs.utk.edu/~browne/perftools-review http://www.tc.cornell.edu/UserDoc/Software/PTools/vampir/
*SGI Implementation of MPI for the Power Challenge
Version 1.0 of SGI's implementation of MPI was released in September and is available for free. You can order it by using the marketing code SC4-PCAS-1.0.
*Active Message Implementation of MPI from Berkeley
The goal of this work is to perform a systematic study on building portable and efficient message passing layers using Active Messages. Active Messages is a simple and efficient communication architecture that exposes the hardware capability and performance. Our approach is to view Active Messages as an instruction set for communication and compose various message passing functions using Active Message primitives.
*MPI for the Fujitsu AP1000
David Sitsky from Australian National University has developed an MPI implementation for the Fujitsu AP1000.
*CRI/EPCC MPI for T3D
Edinburgh Parallel Computing Center (EPCC) has worked with Cray Research, Inc. (CRI) to develop an implementation of MPI for the Cray T3D.
*MPIF for the IBM SP1/2
MPIF is an efficient implementation of MPI from IBM for the IBM SP1/2.
Last modified: Fri Oct 20 11:19:42 CDT 2000 -- Accesses since Thu May 18, 1995: