TFCC Newsletter
Volume 2, Number 1, April, 2000
|
|
TFCC Newsletter |
|
In case you want to access slides of plenary talk and other key speeches, please access: http://www.dgs.monash.edu.au/~rajkumar/tfcc/IWCC99/
URL: http://www.tu-chemnitz.de/informatik/RA/cluster2000/index.htm
URL: http://www.extreme.indiana.edu/java00
URL: http://orca.st.usm.edu/marcin/mp/cfp/DSOEEA/cfp.html
They also have posted the presentations from the prior "event" -- on that page, click on the obvious thing in the upper right, under "Past Events."
Now we would like to invite interested TFCC members to join this project and let us know which part of project you are interested in. Those having comments and suggestions on the project, please let us know. URL: http://www.ComputePower.com/
Dr Kenji Takeda, School of Engineering Sciences, University of Southampton, UK
Editor, Windows Clusters Resource Centre @ www.windowsclusters.org
The remarkable advances in Personal Computer (PC) technology are creating a major upheaval in high performance computing that is hard to ignore. The Beowulf project has helped spearhead the cluster revolution, with the adoption of an open-source operating system (typically Linux) as a key to its success, and continues to drive the cluster community forward at a tremendous pace.
While Linux-based systems are suitable for many environments, the allure of tapping into the huge installed base of Microsoft Windows PCs is tremendous. The success of the Berkeley SETI @home(Search for Extraterretrial Intelligence) project, which packages its radar signal analysis software as a Windows 95/98/NT screensaver, has been tremendous and demonstrates the capability of distributed computing on a global scale. If this type of metacomputing can be even partially realised within a given organisation, be it a major corporation or small business, the pooled computing resource would be impressive. On a more modest scale, even a handful of office PCs can provide enough parallel computing power to significantly improve throughput and make a difference to companies who might not otherwise be able to afford a dedicated compute server.
Decisions regarding operating systems go well beyond the scope of this article and must take into account management and strategic issues. For PC clusters, the OS choice is primarily Linux versus Windows NT/2000. Of course, there are vociferous proponents for both camps, making the choice difficult in the light of much subjective information. The important thing to bear in mind is that there is a choice, and circumstances will often determine the outcome before the technical issues become of paramount importance. And note, for parallel computing it is not the operating system that is most important for performance, but the compiler and communications infrastructure (eg: MPI over fast Ethernet) that will typically determine application efficiency.
The development of Microsoft Windows-based clusters has lagged behind the Beowulf project and its descendants, but is becoming more prevalent. The NCSA NT Supercluster, part of the Illinois Fast Messages/High Performance Virtual Machines project, and the Cornell Theory Centre's AC3 cluster are two prime examples of how Windows Clusters can provide the highest level of scalability for traditional supercomputer applications. Both of these machines have 256+ nodes, use dedicated high performance networking and run large scientific and engineering applications ported from UNIX-based systems to Windows NT. At the other end of the scale is the use of existing office PCs for compute-intensive tasks. For example, codes such as PAFEC-FE Vibroacoustic and Fluent on NT can run efficiently across ethernet networks and open up possibilites for small and medium sized businesses to exploit their IT assets to the full.
The last 2-3 years has seen the emergence of much more hardware and software for cluster computing under Microsoft Windows. There are now commercially supported versions of MPI for Windows NT from MPI Pro Technology Inc and Genias, as well as several other freely available implementations including an official port of MPICH from Argonne National Laboratory. Established cluster management software such as LSF and Condor is now available for Windows, and with Terminal Services now built into Windows 2000 Advanced Server, it is easier to remotely access and centrally manage a cluster. The pieces are in place to be able to make cost-effective use of Windows PCs for compute-intensive parallel applications from the desktop to supercluster level.
To help provide a single source of information and discussion forum for Window Clusters users, the University of Southampton has launched the Windows Clusters Resource Centre, with support from Microsoft Research. It is hoped that this Resource Centre will be useful to the community as a whole and will grow as contributions from other groups are added.
