Gigantic clusters: Where are they and what are they doing?

COLLOSAL CLUSTERS AT ALLIANCE CENTERS

The Albuquerque High Performance Computing Center at the University of New Mexico has long been a proponent of colossal clusters. The AHPCC and the National Computational Science Alliance (the Alliance), comprising more than 50 academic, government, and industry research partners from across the US, have formed a partnership that the National Science Foundation funds. The Alliance, which wants to provide an advanced computational infrastructure, is running a 128-processor Linux SuperCluster with Myrinet (Roadrunner) from Alta Technologies using dual Intel 450-MHz nodes, each with 512 Mbytes of RAM. The AHPCC is acquiring a 512-processor Linux SuperCluster known as Los Lobos, reports David Bader of the University of New Mexico. The Alliance intends to make Los Lobos the largest open-production Linux supercluster geared to the research community.

Los Lobos uses dual Intel 733-MHz IA-32 processor Netfinity 4500R nodes; 1 Gbyte of memory per node; 1 Tbyte of SSA RAID; and 2 Tbytes of tertiary storage (tape robot) to deliver a peak theoretical performance of 375 Gflops. The high-performance interconnect network between the cluster nodes is Myricom’s Myrinet, providing speeds exceeding 1 Gbits per second, which is comparable to the fastest interconnects in today’s traditional supercomputers. The system maximizes the computing power per square foot on an Intel-based platform. This thin server is designed to deliver the highest computing power per square foot on Intel-based platforms and is one of the industry’s most complete rackoptimized product lines for Linux, Windows, and Novell servers, according to IBM. Patricia Kovatch, High Performance Computing Systems Group manager, thinks Linux clusters are compelling for several reasons, especially for the cost-performance ratio. Furthermore, the cost is much less than buying a traditional supercomputer, and the performance rivals one. Another benefit for applications folks is the ease of porting their applications to Linux from other Unix environments. The prospects for the near future look even more momentous: “Multiple multiple-terascale Linux-based superclusters will be built in the next year with a 10-terascale or better Linux supercluster highly likely in about a year,” says Frank Gilfeather, Executive Director of High Performance Computing.

Another Alliance partner, the National Center for Supercomputing Applications at the University of Illinois at Urbana- Champaign, is running a 128-processor dual-Pentium-III Xeon, 550-MHz, 1-Gbyte RAM, and Myrinet network technology on an NT operating system.