The cluster will be used by a varied group of scientists from two top research institutes of the University of Utrecht. Scientists from the Astronomical Institute will use the cluster for research concerning the evolution of stars and binary star constellations and their interaction with the interstellar medium. This will later be extended with Magneto Hydrodynamic problems associated with star atmospheres and more exotic phenomena like quasars and black holes.

Scientists from the Institute for Marine and Atmospheric research Utrecht (IMAU) will use the cluster for the modelling of ocean currents such as the Gulfstream, flows in the atmosphere, the transport of chemical components such as ozone, the movement of sandbanks in coastal areas and the behaviour of ice caps and glaciers. In all these areas, a finer detail of representation of physical processes is becoming possible through the availability of the cluster.

Prof. dr ir Henk Dijkstra, professor of Dynamical Oceanography and director of IMAU, said: "We are very excited to be one of the first to use the new InfiniBand technology to boost the computing power available to the scientists at IMAU. We chose the combination of products and services from ClusterVision, Fabric Networks and Intel because of their track-record and proven expertise in High Performance Computing."

The cluster will also be used by the High Performance Computing Group of the university. Dr Aad van der Steen, leader of the research group said: "We are very pleased with the new cluster because the very low latency and the extremely high bandwidth of InfiniBand allows us to study new application areas that traditionally were hard to tackle such as combinatorial and Data Mining problems in Bioinformatics."

InfiniBand is a new network technology based on an open standard I/O architecture developed by a consortium of some of the world's largest computer companies, including Intel, Hewlett-Packard, IBM and Microsoft. InfiniBand has a bandwidth that is almost 10 times higher and a latency that is almost 10 times lower than that of traditional Gigabit Ethernet. It also reduces communications overhead by as much as 50% of server CPU cycles.

ClusterVision will supply the cluster with the MVAPICH parallel programming libraries. MVAPICH is an MPI implementation for InfiniBand developed by the research group of Prof. Dhabaleswar K. Panda, professor of Computer and Information Science at Ohio State University in the USA. "We are using the Fabric Networks solution in our laboratory today and have achieved impressive results. Our testing with the MVAPICH package is showing latencies for MPI applications of 6.8 microseconds for small message sizes and bandwidths of over 850 MB/s for large messages," said professor Panda.

The cluster will consist of 33 SuperMicro rack-mount servers with 66 Intel Xeon CPUs at 2.8 GHz and a total of 66 GB of memory. Each server will have a Fabric Networks HCA-X InfiniBand PCI-X card connected to a Fabric Networks FabNet 3200 32-port InfiniBand switch. The cluster will be running the Linux-based ClusterVisionOS cluster operating system and software environment which includes all software required to effectively use and manage the cluster.

About Linux Supercomputer Clusters

ClusterVision's clustering technology provides an alternative to traditional supercomputing by connecting multiple computers to form a unified and powerful computing system. This type of computer cluster is also called "Beowulf Cluster". The clusters are built from high-quality commodity computer components, Linux and other proven open source software technology. Clusters can therefore match the performance and stability of traditional supercomputers for a fraction of the cost.

About ClusterVision

ClusterVision are specialists in the design, implementation and support of large-scale compute, storage and database clusters. The technical team at ClusterVision has designed, implemented and supported some of the largest and most complex compute and storage clusters in the UK and the Netherlands. Customers include Rutherford Appleton Laboratory, Natural History Museum, Imperial College and Manchester University in the UK, and Delft Hydraulics, TNO and the University of Leiden in the Netherlands. ClusterVision has offices in Amsterdam and London.

About Fabric Networks Corporation

Fabric Networks is the leading provider of Switched Fabric Networking solutions for Enterprise Data Center high availability computing and High-Performance Computing server clusters. Created from the strategic merger of InfiniSwitch and Lane15 Software, Fabric Networks offers a legacy of excellence in both switched fabric hardware and fabric management software. Fabric Networks is an IBM ServerProven Partner and an OEM partner with other leading server manufacturers. For more information visit:

For high-resolution pictures
Contact: press@clustervision.com

For more information

Dr ir Matthijs van Leeuwen
ClusterVision BV
J.C. Beetslaan 155
2131 AL Hoofddorp
Netherlands
Tel NL: +31 23 567 7510
Tel UK: +44 7966 012215
Fax: +31 23 567 7519
m.vanleeuwen@clustervision.com
http://www.clustervision.com