RealTime IT News

Supercomputing Giants Drum Up Show Buzz

Leading makers of computers that perform massively parallel transactions are gearing up for the Supercomputing 2003 conference in Phoenix (Nov. 17-21) with a flurry of news announcements this week.

IBM HP, Sun Microsystems, Dell, SGI and Cray make up the bulk of vendors who craft machines capable of making massive amounts of computing power available to customers.

All of the vendors are locked in fierce battle to garner new customers and more market share in a segment commonly known as high performance computing, or supercomputing, which employ grid and clustering technologies to serve multiple systems linked together to act as a single engine.

The fact that the machines are so large and employ so many servers and clusters makes the contracts with major institutions that much more valuable. IDC holds that the supercomputing market is worth around $5 to $6 billion and grows about 10 percent per year.

Issued from the company's Deep Computing division, IBM and Corning inked a contract with the government, unleashed a diskless InfiniBand supercomputer for Mississippi State University and used AMD's Opteron chips in a cluster product.

Meanwhile, HP teamed with BAE Systems and three institutions to build an advanced global computing grid for aerospace and defense design; Dell issued a 34-node cluster to support Princeton University student and faculty research; and a Sun executive discussed his company's demonstrations at the show.

Earlier this week, Big Blue joined forces with Corning in an agreement to provide the U.S. Department of Energy and the National Nuclear Security a Administration (NNSA) on a $20 million project to develop optically switched interconnects for supercomputers. The goal is to replace copper cables and electronic switches with optical technology to boost network bandwidth by 50 times.

IBM said this will address a challenge in the design of high-performance computer systems to match advances in microprocessor performance with advances in data transfer performance, noting that U.S. government agencies expect to eventually scale supercomputers to thousands of nodes with interconnect bandwidth of tens of gigabytes per second per node.

Corning's Science & Technology Division will develop a prototype of an optically switched interconnect. IBM Research Labs in the U.S. and Switzerland will provide the system's electronic control and monitoring circuitry and will assist with the integration of the optical interconnect modules provided by Corning to deliver a prototype system within 30 months.

Thursday IBM inked a major contract with Mississippi State University (MSU) to install what will be the largest supercomputer powered by InfiniBand connection technology. Financial terms of the deal were not revealed, although an IBM spokesman confirmed that it was multi-million deal.

Dubbed "Maverick," the massive cluster employs InfiniBand from Voltaire to facilitate communications between 192 eServer Intel-based xSeries 335 servers, which will perform such tasks as testing the "crash-worthiness" of new automobiles to make cars safer under the auspices of MSU's Engineering Research Center (ERC).

Other Maverick projects will include a "virtual wind tunnel," which tests the aerodynamics of automobiles, aircraft, and ships to increase performance and fuel efficiency. The ERC is also researching how to predict the formation of ice on aircraft wings to make flight safer.

ERC will shepherd Maverick through a number of tests, but one of the attractive selling points in this deal is that Maverick can reduce the time required to complete a test, saving companies thousands of dollars. The combination of eServers and InfiniBand gave the ERC a 400 percent increase in processing power over its current cluster, said David Marcum, Director of the ERC SimCenter.

Dave Turek, vice president of IBM's Deep Computing unit, said one of his group's goals is to extend the appeal of supercomputers for segments beyond scientific organizations to financial services or digital media concerns. But, he conceded, those companies need to be assured that those solutions are practical, which is why they will wait to see if clusters are made practical in scientific environments.

IBM, which has embraced Intel's Itanium architecture as well as its own Power chip line for servers, unveiled a pre-packaged super cluster based on AMD's Opteron. Turek told internetnews.com that IBM had little choice in the matter. "Customers demanded it," he said, noting that Opteron is quite a natural migration to 32-bit and 64-bit systems from x86.

"You get to choose the rate and pace to move to 64-bit," Turek said. "There's no forklift upgrade like there is with Itanium and the applications are easy to deploy. The price/performance ratio makes it a really, really good system."

The new eServer 1350 can run 32- and 64-bit applications at the same time. This product has a core for the new cluster in the form of the eServer 325, which employs 48 Opteron 2GHz processors. The eServer Cluster 1350 combines the eServer 325 to form a system with more management and storage systems to provide customers more choice in setting up Linux clusters.

Like Maverick, the eServer 325 is designed for scientific and technical computing customers. While it doesn't leverage InfiniBand, Turek said its high computing capability and integrated memory controller alleviate the usual bottleneck issues. The eServer Cluster 1350 can be built with any combination of IBM BladeCenter systems, x335 and x345 systems.

Other supercomputing news

HP, largely acknowledged as the HPC leader a year ago, Wednesday teamed with BAE Systems on a supercomputing venture for aerospace and defense design.

The collaboration will be based on open standards like the Globus Toolkit and the Open Grid Services Architecture, which combines Grid technology with Web services. HP and BAE say they will develop improved security measures for sharing Grid-enabled applications and data.

As the electronic systems used in the aerospace and defense sectors become increasingly elaborate, the process of designing, visualizing and simulating the operation of these systems becomes critical, HP and BAE said.

"Aerospace is on the leading edge of organizations that collaborate one day and compete the next," Sara Murphy, HP's marketing manager for Grid computing told internetnews.com sister site Gridcomputingplanet.com.. "This is a wonderful opportunity to understand the core issues of collaboration and virtual organizations. ... It's a good proving ground for Grid's commercial possibilities."

In still other high-performance computing news is supplying a high-performance computing cluster (HPCC) for anything from engineering and physics to the study of the brain, mind and behavior at Princeton University. The 34-node cluster will serve as a teaching tool for classes and workshops in order to facilitate better training in computational techniques and theory and serve as a testing ground for HPCC planning prior to investing resources.

Dell said that because its HPCCs are based on open standards and powered by the open-source Linux operating system, they typically operate at a lower cost than supercomputers based on proprietary technology. The cluster is comprised of 34 PowerEdge 2650 servers, each with two 2.4-gigahertz Intel processors running the Red Hat Linux OS.

Let the Sun shine in

Meanwhile, building on the momentum of a $50 million high-performance computing contract with DARPA this past summer, Sun is sticking to its x86 guns, according to Steve Perrenod, Group Manager, HPTC at Sun.

The executive said the company, which IDC said grew its HPC offerings 43 percent versus 30 percent for the overall market from 2002-2003, will sample from its supercomputing, grid, and storage-oriented itinerary in Phoenix at the show that will include 20 demos, 14 racks and 11 partners.

In an unusual move, Perrenod said Sun will work with the University of California at San Diego to build a cluster live on the show floor with nodes that can be ordered at random anywhere and in any software stack. It will employ Sun's Grid Engine software to perform tasks while the cluster is being constructed. Demos will treat such topics as visualization, data storage, cluster and grid computing, including the Sun Fire V60x Compute Grid Rack System based on SPARC/Solaris and x86/Linux architectures.

Sun will also demonstrate its progress in deploying Linux-clustered environments, as well as a high performance computing SAN solution employing myriad heterogeneous technologies. Perrenod also expressed confidence that Sun will regain lost traction on the all-important Top 500 Supercomputer list, which is unveiled annually at the show.

"Our ranking has been down the last couple of lists, but we believe we will be show progress with our large Linux clusters," Perrenod told internetnews.com. The vendor will also demo high-end superclusters that use the company's Sun Firelink interconnect technology. He also predicted the rise of InfiniBand and the ability to employ optical networking as interconnect technologies to be some trends to watch out for.

"We're seeing more diversity in the solutions and architectures using optical technology," Perrenod said. "The price is still a little high but the performance is there. I think a very big trend over the next few years will be the notion of driving optics into computer system designs. It's a great opportunity."