Hewlett-Packard to Claim "Fastest Linux Cluster" Crown with Itanium-based Beowulf for DOE
The US Department of Energy's Pacific Northwest National Laboratory has ordered a $24.5 million, 1,400 processor Linux cluster that, when completed, will be the most powerful Linux-based system in the world, said Martin Fink, HP's general manager for Linux systems operation.
HP said in an announcement that the new cluster is expected to achieve 8.3 teraflops. By comparison, the current head of the Top500.org list of the world's fastest computers is IBM's ASCI White, which reaches about 7.2 teraflops.
The cluster, to be installed beginning around the middle of this year and concluding next year, will be used for environmental and energy research, such as predicting the reactions of uranium wastes under the Department of Energy's Hanford Nuclear Reservation, said Dr. David Dixon, associate director of theory, modeling and simulation at PNNL's William R. Wiley Environmental Molecular Sciences Laboratory (EMSL).
The processors will be Intel's forthcoming "McKinley" and "Madison" versions of the Itanium 64-bit processor and will use Quadrics QsNet interconnects, which boast a five microsecond interconnect latency, Fink said. "We've run our codes on Quadrics interconnect and they scale phenomenally", said Scott Studham, group leader of computer operations at EMSL. The first nodes to be installed will be deployed with QsNet 1, then upgraded to the higher-capacity QsNet 2, and later nodes will be preinstalled with QsNet 2, Studham said.
HP will be installing Intel's compilers for Itanium and MSC Linux clustering software, Fink said. In addition, EMSL will deploy its own job scheduling software developed in-house, Dixon said. In making the choice of Linux, EMSL went with what HP advocated in their winning bid, Dixon said. "The Linux part wasn't the dominant part, it was the hardware performance."
Don Marti is technical editor of Linux Journal.