Fastest Computer Cluster Architecture in the World Unveiled in Dresden
[29. Juni 2006]
Jülich, 29 June 2006 - Research Centre Jülich is working on the next generation of high-performance cluster computers in conjunction with companies from Germany and USA. A prototype is now being unveiled in front of an expert audience at the International Supercomputing Conference ISC 2006 in Dresden. With this system, a new cost-efficient power class has been realised, representing a milestone in cluster computing. The precise combination of the components optimises communication between the high-performance processing units.
Cluster computers are built modularly out of ordinary PC components and fast networks. They are suitable for a wide range of simulation problems and currently represent a large proportion of the most powerful computers in the world. The decisive factor is the fast transmission of signals across the network, the "backbone" of the computer.
"The Jülich Linux cluster JULI connects the fastest processors with the most efficient network at the moment over standardised interfaces", says Dr. Norbert Eicker of Research Centre Jülich, who is coordinating the development of the largest German supercomputer centre, the John von Neumann Institute in Jülich. "In addition, we also have the cluster operating system, ParaStation, which controls the way that the components work together." ParaStation was developed by the Munich-based company ParTec and the experts in Jülich. The processors in IBM's JS21 blades are connected over the "InfiniPath" network developed by the US company, Qlogic. Adjustment of the interfaces occurred in a record time of only four months through the IBM Research Laboratory in Böblingen. This demonstrates just how powerful IBM's Technology Collaboration Solution actually is.
The IBM JS21 blades, which are equipped with two PowerPC 970MP dual-core processors with a top speed of 2.5 GHz, are three times faster than any of the previous blade generations. InfiniPath provides a particularly high bandwidth even for small reports and has a latency time of less than 2 µs. ParaStation ensures high availability, performance and reliability of the system and provides both administrators and users with a comprehensive overview of the sequence of operations performed by the system through the ParaStation grid monitor.
JULI will be put into operation as a pilot system in order to advance its development. Together with the IBM Blue Gene system, JUBL (46 teraflops), which went into operation in March of this year, JULI is the next important step for Research Centre Jülich on the way to becoming Europe's petaflop-range supercomputer complex.
Science Journalist, Research Centre Jülich
Tel. +49 2461 61-4771, Fax +49 2461 61-4666