Virginia Tech Edging Towards Finish Line In The Supercomputer Race

Steger acknowledged that Virginia Tech is in a supercomputing race. And the finish line is a moving target. It may come at the end of this week, but no later than Oct. 1.

One of Virginia Tech’s engineering faculty members, Srinidhi Varadarajan, and his colleagues are building a groundbreaking, low-cost, very high-speed terascale supercomputer. This supercomputer will be one of international prominence, yet based upon a homegrown cluster using off-the-shelf industry components.

For the past three months, the group has been working around the clock, seven days a week to complete this project by Oct. 1.

They are tying together 1,100 of Apple’s new G5 desktop computers to achieve top-dollar performance. Tech’s strategy is a departure from monolithic mainframe supercomputing and is using a less-expensive but reliable configuration.

Virginia Tech’s new solution for creating supercomputing clusters will cost about one/tenth of what most supercomputers cost. As such, this supercomputer should serve as a model that will allow other universities, who are not in the supercomputer game, to own their own facilities.

Virginia Tech’s costs – spread out over a five year period – will be $5.2 million. Put in perspective, that’s a cost of $1 million a year. And last year, Tech’s College of Engineering alone attracted more than $100 million dollars in research expenditures. The new supercomputer is expected to bring to Virginia Tech numerous big science research projects that it did not have the resources for previously.

Varadarajan is also the developer of “Déjà vu,” a software package that brings stability to large clusters. He is incorporating this software into the 1100-node cluster, making Tech’s supercomputer the only one in the world currently operating with this software solution to a 20-year old reliability problem. The National Science Foundation (NSF) supported Varadarajan’s work in the development of Déjà vu.

The reason Virginia Tech is racing to assemble this supercomputer by Oct. 1 is this date is the deadline for the next contest that will rank the world's top supercomputers. “We believe we will have one of the top-ranked supercomputing facilities in the world when the project is completed,” Steger reiterated.

Virginia Tech’s partners for building this supercomputer in less than three months are Apple, Mellanox, Cisco, and Liebert. Mellanox is the leading provider of the InfiniBand semiconductor technology, the primary communications fabric, drivers, cards, and switches for the project. Cisco’s Gigabit Ethernet switches were the choice for the secondary communications fabric to interconnect the cluster. Cisco provided a significant educational discount to support the project. Liebert, a division of Emerson Network Power, known for its comprehensive range of protection systems for sensitive electronics, provided the cooling system.

Virginia Tech students have volunteered hundreds of hours to help set up 19.25 tons of computers, routers, and other equipment.

When Virginia Tech is successful in launching its supercomputer later this month, “We will be among the few universities that will be able to effectively compete for the Cyberinfrastructure funds. This proposed National Science Foundation program is scheduled to have a $1 billion annual budget,” Steger added.

Virginia Tech is already a participant in the National Lambda Rail, a consortium of leading U.S. research universities and private sector technology companies. National Lambda Rail has launched a project that is a national facilities-based approach for optical networking and network research. This initiative is tying together multiple clusters via very high performance networks for computational science – big science – problems. Virginia Tech plays a leading role in this project.

More information about the Terascale supercomputing project can be found at: http://computing.vt.edu/research_computing/terascaleMore information about the National Lambda Rail project can be found at: http://www.nationallambdarail.org