A NEW SUPERCOMPUTER FACILITY AT PARALLAB

The University of Bergen recently acquired a new large computer facility. The primary objective of the new supercomputer is to serve as a high-throughput platform for a wide variety of computationally intensive scientific applications. The target user groups are bio-informatics, computational physics, computational chemistry, and climate sciences. The system complements the IBM Regatta p690 (shared-memory) installation that is being operated by Parallab/BCCS for UiB since early 2002.

The new facility is an IBM e1350 cluster solution that consists of 86 e326 compute nodes, each with 2 AMD/opteron model 250 (2.4 Ghz) processors and 80 GB local disk. The machine includes 'thin' nodes (2 GB memory), 'thick' nodes (4 GB memory) and 'fat' nodes (8 GB memory) to accommodate the range of applications in a cost-effective way. The operating system is Linux. The node interconnect is Gigabit Ethernet.

The aggregate numbers for the cluster are 825 Gflops/s (billion operations per second) theoretical peak performance, 258 GB main memory, and 14.8 TB disk storage (incl. an 8 TB external disk solution). The performance of the machine is roughly equal to 170 modern PCs.

Special attention has been given to accommodating scientific applications that require high I/O performance. Therefore, the system also includes three x346 nodes (2-cpu EM64T Xeon, 4 GB RAM, 144 GB disk) that are dedicated to provide high bandwidth between the e326 compute nodes and the 8 TB external disk. The installation is integrated in the current HPC environment at Parallab/BCCS that includes for example 80 Tbyte secondary storage (tape robot).

In the acquisition process, the chosen solution competed with other cluster solutions, one of which had a significantly higher aggregate theoretical peak performance of nearly 2 Tflops. The observed sustained performance results during the benchmarking process and feedback from the target user groups however decided in favour of the configuration as mentioned above.

The supercomputer facility is installed at Høyteknologisenteret in Bergen and managed and operated by staff of Parallab/BCCS. The machine came alive in the week of November 15-19. The configuration is being verified by BCCS staff and selected users. The system will be opened up to the targetted research community early December.

On December 9, a special seminar is organized at the Høyteknologisenteret to inaugurate the machine. See below for more details.

For more information about this cluster, please contact tel. 55584314 or email adm@bccs.uib.no (or jak@bccs.uib.no).

An overview of all supercomputing facilities currently operated by BCCS can be found here.

Seminar
Big Auditorium (rom 2144), Høyteknologisenteret, Datablokk
Thursday, December 9 2004, 12:30-14:15
Open for all

 12:00 Lunch (for invitees only)
 12:35 Prorektor Rune Nilsen, Universitetet i Bergen
 12:45 Perspektiv på utviklingen av beregningsverktøy
       Professor Petter Bjørstad, Institutt for Informatikk
 13:00 Presentasjon av det nye anlegget
       Overingeniør Jan-Frode Myklebust, Avdeling for Beregningsvitenskap (BCCS)
 13:15 Simulations of phase transition dynamics and practical implications
       Professor Bjørn Kvamme, Institutt for Fysikk og Teknologi
 13:30 Computational Biology
       Gruppeleder Nathalie Reuter, Avdeling for Beregningsvitenskap (BCCS)
 13:45 Scientific visualization
       Dr. Christopher Giertsen, C. Michelsen Research
 14:00 Numerisk klimamodellering
       Gruppeleder Nils Gunnar Kvamstø, Bjerknessenteret for Klimaforskning
 14:15 Besøk i maskinrom for interesserte