Supercomputers are raising campus research profiles

A powerful new tool will be arriving in September at the University of Massachusetts Dartmouth campus, a supercomputer with the capacity of 4,000 desktop machines.
In recent years acquiring such technology has become a goal for many universities and research facilities around the globe. UMass Dartmouth is only the second institution in the region to make the jump. Brown University unveiled its new supercomputer last November. The University of Rhode Island is also eyeing such technology as well.
When the new equipment is up and running at UMass Dartmouth – which should happen in November – the school could become a leader in technological and scientific research.
“It’s raising our engineering and computational science profile,” said Robert Fisher, a physics professor at the school. “This will truly be a world-class resource for our scientists. By enhancing our computational abilities, we’ll be able to attract more state and federal grant funding to the university.”
Individual researchers have access to other institutions’ supercomputers, “But you have to schedule a time to use that kind of equipment, and it could take as much as a year,” said Sigal Gottlieb, a mathematics professor at the university. “That can really hold up your research.
“One reason we’re so excited about our proposal is the range of applications we’re looking at. We have people talking about simulating supernovas and crystal growth,” she said. “We’re looking at ocean modeling and atmospheric modeling. One of our faculty members is doing tear-film modeling, an important project in biology.”
A supercomputer is a custom-made, one-of-a-kind machine capable of extreme high-speed calculations. They’re used by researchers to model complex phenomena and conduct numerical experiments related to everything from quantum physics to weather forecasting and economics. Most process terabytes of data – equal to one trillion bytes – in seconds, but some are even faster. The Roadrunner, an American military supercomputer located in a laboratory in Los Alamos, N.M., has set a milestone by processing more than one quadrillion calculations per second.
Brown’s multimillion-dollar machine is not quite as fast but can perform a still impressive 14 trillion calculations per second at its peak speed of 14 teraflops – 50 times faster than what was previously possible at Brown. A group of researchers mapping Mars can now process images from NASA in one day instead of three months using Brown’s machine. In recent years acquiring such technology has become a goal for many universities and research facilities around the globe.
URI, for example, is considering how best to get involved.
“We’re in the exploration stage at this point,” said Garry Bozylinsky, URI’s vice provost of information and technical services. “The question is whether it makes sense for us to get something on our own, or connect with Brown. Either way, we’ll definitely partner with Brown.” If URI got its own, “It would be something that would allow us to combine [it with Brown’s], and make them work as one machine,” he said.
“We’ll need a high-speed fiber network that will allow us to connect with the Providence area. That will probably happen next year,” Bozylinsky said.
Rather than a crosstown rivalry shaping up among computer scientists at the region’s various schools, there is more of an interest in collaboration, as URI’s approach suggests. Faculty at UMass, however, admit there is sometimes a feeling of global competition.
“It’s a matter of keeping up with computational science,” said Gottlieb. “Whenever we’re talking about high-performance computing, the fastest are only available in a handful of places. They’re typically called Track 2 systems. We’re getting what’s called a Track 3 [the next-fastest system]. If you want people developing codes and algorithms for these systems, you have to make them available to people in the field. You can’t train scientists for this work on a desktop. You need the Track 3.”
Ironically, the push for supercomputers at universities has been spurred by devices that many educators scorn as wastes of time – the Sony PlayStation and other video game systems. Over the past decade, competition in the entertainment industry has pushed companies to invest in research, resulting in faster technology. Supercomputers get their boost from graphics processing units – or GPUs – which were first developed to make video games visually exciting. “The world of scientific research is a niche market, so the development of the GPU has really been driven by the entertainment market,” said Fisher. “A single GPU is about 50 times more powerful than the traditional central processing unit.”
Proof of the game console’s strength can be found right on the UMass Dartmouth campus. Last year, faculty members Gaurav Khanna and Chris Poulin made headlines when they linked eight consoles together and created a small supercomputer, which they used to predict the properties of gravitational waves generated by black holes. Their do-it-yourself guide is posted online at
www.ps3cluster.org.
The GPU has also made supercomputers affordable to more institutions. The price tag of the UMass system, for example, is expected to be about a half million dollars; if the central processing units favored in the past were used, the cost would be in the millions.
Like most of today’s supercomputers, the UMass system will actually be a cluster of computers linked together. Each of the three will be equipped with Fermi GPU chips made by NVIDIA. “The technology has only become available in the past few months,” Fisher said. “It’s the same technology that’s used in the Chinese supercomputer that’s now ranked No. 2 on all the supercomputer lists.”
For such a powerful tool, it will be fairly small. The entire system will sit on two rack mounts, each about the height of a person. It will be housed in the university data center in the Claire T. Carney Library. There will be no monitor. Researchers will access the supercomputer via a secure Internet connection.
A $200,000 grant from the National Science Foundation and funds available through the federal economic stimulus law will pay for the UMass supercomputer.
In addition to faculty, the supercomputer will also be used by graduate students. UMass is now developing a doctorate program in computational science, and those in mathematics, engineering, physics and other science and technology fields will have access to the machine as well. •