Turning a cluster of Windows NT machines into a parallel supercomputer

Almere, 23 November 98 A whole generation of engineers is growing up with knowledge from Windows NT. Also NT based computers are entering the computing arena, replacing Unix based workstations. When supercomputer type of performance was needed, there was no other alternative than to turn to Unix. Not so any more. In the European project WINPAR tools for parallel processing are ported to the Windows NT environment. With these tools, a cluster of NT machines can be turned into parallel supercomputers as powerful as their Unix counterparts. A first product, a commercial version of WMPI, implementing the standard message passing interface has recently hit the market. And how, declared best in functionality and performance by an independent report.

"PaTENT MPI 4.0" the new product was dubbed, an acronym, for "Parallel Tools Environment on NT". Acronyms are a typical European way of naming products. The quality of the product is however world class. An independent evaluation of MPI on NT products shows that it stands out against implementations and products that originate from the USA.

The Message Passing Interface, MPI, is an international standard for parallel computing. With MPI, a bunch of computers can be turned into one big parallel machine that works on one computing application as a single large machine would do.

PaTENT MPI is based on WMPI from the Portuguese Instituto Superior de Engenharia de Coimbra.Within the Winpar project, the software was turned from an outstanding research result into a quality commercial product by German company Genias Software.

PaTENT MPI can be used on any Windows NT platform or cluster. There are interfaces to Microsoft Visual C++ and Digital Fortran. The product adheres to the MPI 1.1 standard and is "MPI 2 ready". Hence, applications written in these languages can make use of parallel processing when a company or institute has several workstations available. Also servers with more processors can be used.

For example, the software has been used as part of a thread rolling application with the MARC Analysis software. In this simulation, the process of making a metal thread from a cylinder by pressing it between rotating pressing dies is calculated. The application was tested on 1,2 and 4 processors. On 2 processors, the application ran 1.9 times as fast as on a single processor. On 4 processors, this factor was 3.0.

And this is just the beginning of power computing on NT. The WINPAR project, Esprit nr 23516, promises us PVM on NT, TRAPPER, AUGUR, DIWIDE and MODARCH: a complete suite of tools. They will be brought to the market by Genias and Simulog. Other partners in the project are MTA SZTAKI in Hungary, DASH in England, TRANSVALOR in France, GMD in Germany, the University of Coimbra in Portugal, The University of Vienna in Austria and the University of Miskolc in Hungary.

For more information check in at the WINPAR web site
The MPI on NT comparison is available from Mark Baker from the University of Portsmouth.

Ad Emmen