NASHVILLE, Tenn. – What do Paul Sheldon, who studies the elementary particles, Jason Moore, who analyzes high-dimensional genetic data, and Ron Schrimpf, who investigates the effects of radiation on space electronics, have in common?
They are all serious "number crunchers" – researchers whose work requires the processing of such large amounts of data that they need the services of a supercomputer. A supercomputer is simply a computer that performs at or near the highest processing rate available. Although computers keep getting faster, the three researchers predict that more and more of their colleagues in a variety of disciplines will join them in the ranks of "high performance computation" users over the next decade.
"High performance computation is going to become hugely important to all fields [of science]," says Paul Sheldon, associate professor of physics.
That conviction, which is shared by Moore, Ingram Associate Professor of Cancer Research, and Schrimpf, professor of electrical engineering, led the three to spearhead the effort to create a state-of-the-art supercomputer center on the Vanderbilt campus, one that would be available to researchers from all disciplines. Now, their vision will become a reality with an $8.3 million grant from Vanderbilt’s Academic Venture Capital Fund.
"We’ve defined three general categories of users," says Moore. "People like us who need this stuff and use it on a daily basis. Then there’s sort of a second class of users that are doing computing and would like to get into parallel computing but have never had the opportunity or the resources to do it. And then there’s a third class of users that this isn’t even on their radar screen – they just have no clue that this is something that can help them."
In order to cater to all three types of users, the money for the new center will be used not just on computer equipment but also to set up a training and outreach center designed to inform supercomputer novices about the advantages it offers and to help new supercomputer users adapt their applications to run on the more powerful machines.
Plans also call for the creation of a "web portal" that will make it easy for users to submit jobs and get results over the Internet, rather than having to trudge back and forth to the Hill Center, where the new Supercomputer Center (SCC) will be located.
"Other similar high-performance computing centers located at places like Louisiana State University, Cornell and the University of Florida are devoted to a single program or discipline," says Sheldon. "No one else that we know about has tried to implement the multidisciplinary idea." Schrimpf adds, "It should increase our visibility and profile among other researchers worldwide in a way that we would never be able to do otherwise."
According to the supercomputing missionaries, the time for high performance computing to enter the academic mainstream has definitely come. Thirty years ago, supercomputers were massive, bulky, custom-built and so expensive that only government agencies, multinational corporations and the largest research universities could afford to own them. At first, supercomputers were designed around single processors, which were pushed to operate at ever faster and faster speeds. About 15 years ago, however, a new architecture was introduced: machines that employ multiple processors running at the same time called parallel processors. The first of these were also custom designed, but the increasing power and decreasing cost of processors mass-produced for personal computers led inevitably to the idea of strapping together hundreds to thousands of these off-the-shelf units to create "supercomputer clusters" that can provide supercomputer power at much lower costs.
Vanderbilt’s SCC will be built around VAMPIRE – the VAnderbilt Multi Processor Integrated Research Engine – the 200-node cluster created three years ago that has served as a research tool and test bed for the larger and more powerful SCC system. (Each node contains two processors, some dedicated memory and a networking chip.)
Alan Tackett, research assistant professor of physics and technical director of the center, gives an example of the potential advantages of moving from computing on a desktop or scientific workstation to the supercomputer: "We have two neuroscientists who have begun using VAMPIRE. One has an application that was taking two days to run on a normal computer that takes only 20 minutes on the supercomputer. The other neuroscientist has found that he can complete a month’s worth of work in a single day."
Virginia Tech has been in the news recently for building a supercomputer cluster out of 1,100 of the latest Apple Macintosh personal computers. This facility, called the Terascale Cluster, has been measured as the third-fastest supercomputer in the world based on its processing speed. But the Vanderbilt supercomputer advocates are not impressed by these sorts of records. "They may grab headlines, but the way in which they are measured often does not translate into faster processing of real-world applications," says Sheldon.
Furthermore, the race to build the biggest and fastest supercomputer cluster has a definite downside, in their view: The supercomputers are already obsolete before they are even up and running. "It normally takes about a year for the staff to get up to date and to start using it," Tackett explains. "Now you’ve got a 1,000-node cluster that’s already a year old. Training your audience [the research community] takes another couple of years. By then, you’re two or three years down the road and you’ve got a thousand Pentium III boxes … but who cares?"
So Vanderbilt’s SCC will not be trying to set any world speed records. Instead, it will be pursuing a phased development strategy. The plan is to increase the total number of nodes each year by one-third over three years, and then begin to replace the oldest one-third of the nodes each year with the fastest processors available. Each of the new nodes that they install will have more than four times the computing power of an existing node. Therefore, the replacement process will continue to boost the cluster’s performance. In this way, the cluster will benefit from future improvements in computer technology and continue to grow without becoming obsolete.
"We believe that the SCC will have a huge impact on Vanderbilt, on recruiting and retaining top faculty and attracting the best graduate students, because they will have the computing power that they need to tackle the toughest new problems in their respective fields," says Moore.
For more news about Vanderbilt research, visit Exploration, Vanderbilt’s online research magazine, at www.exploration.vanderbilt.edu
Media contact: David F. Salisbury, (615) 343-6803
David.f.Salisbury@vanderbilt.edu