Trestles Supercomputer: Obtaining a Research Edge
Supercomputers give universities a competitive edge in scientific research.
That was the conclusion of a recent study led by Amy Apon, chair of the Computer Science Division in the School of Computing at Clemson University. Before moving to Clemson, Apon was a professor of computer science and computer engineering at the University of Arkansas and founded the Arkansas High Performance Computing Center in 2008.
The Clemson study found that universities with locally available supercomputers were more efficient in producing research in critical fields than universities that lacked supercomputers.
“It is unequivocal that a high-performance computing system will provide an advantage in doing research in several fields,” Apon said. “It’s not uniform across all fields. But for fields where it matters, it matters a lot.”
That’s good news for the U of A.
This spring, the National Science Foundation and the San Diego Supercomputer Center at the University of California, San Diego, transferred ownership of the computer cluster known as “Trestles” to the Arkansas High Performance Computing Center.
The new supercomputer, which became operational in July, more than doubles the center’s computational capacity and allow it to run three times the amount of jobs for campus researchers.
“We are thrilled to acquire a prominent national resource for high performance computing,” said Jeff Pummill, director of outreach for the Arkansas High Performance Computing Center. “Researchers at the University of Arkansas are in a perpetual state of evolution and advancement in their computational needs, and Trestles is known throughout the national high-performance computing community as a ‘high-productivity workhorse.’”
The Arkansas High Performance Computing Center supports research for about 260 users in about 30 academic areas across the campus, including bioinformatics, physics, integrated nanoscience, computational chemistry, computational biomagnetics, materials science and spatial science, among others.
“Trestles will provide much-needed infrastructure for solving tasks such as genome assembly and other data-intensive problems,” Pummill said.
1 MILLION SERVED
Trestles has 256 servers, 16.4 terabytes of memory and a processing speed of 79 teraflops. A teraflop is the measurement of 1 trillion calculations per second, with the “tera” meaning trillion and the “flop” standing for “floating-point operations” per second.
A typical desktop computer has four processing cores, Trestles has 8,192.
U of A’s flagship supercomputer, dubbed Razor, has a processing speed of 77 teraflops from its 4,328 processing cores.
Between them, Razor and Star of Arkansas, the university’s other supercomputer which went offline this summer, performed more than 1 million jobs since the Arkansas High Performance Computing Center’s founding in 2008. The center hit that milestone earlier this year.
“A million is a very substantial number, one that reflects all of the times our users submitted a job and hit ‘go.’ Some of our users submit hundreds jobs at once that won’t take very long to compute, while others can take weeks to finish,” Pummill said. “There are polar extremes to the computations our supercomputers perform on a daily basis.”
Laurent Bellaiche, Distinguished Professor of physics in the J. William Fulbright College of Arts and Sciences. Xu, said supercomputing is essential to his work.
“Without such resource, we could not survive as a computational research group,” he said.
Pummill and David Chaffin, the center’s two interim co-directors, and Pawel Wolinski, its senior Linux cluster administrator, call computational science the “Third Leg of Science,” beside experimental science and theoretical science.
In 2010, the National Science Foundation awarded $2.8 million to the University of California, San Diego to deploy Trestles. Since it went online, the supercomputer has been recognized as the leading science gateway platform in NSF’s eXtreme Digital Network, a collaborative set of computer and storage resources across the United States that scientists can use for advanced computational and data-enabled research.
Pummill became aware last year that Trestles would be decommissioned in 2015. The parties verbally agreed to the transfer last summer.
“We talked with NSF and SDSC about acquiring Trestles,” Pummill said. “We were able to demonstrate the need for such a large system and we had the experience to deploy it. We also had the space, the power and the amount of cooling that was required to maintain Trestles.”
Richard Moore, deputy director of the San Diego Supercomputer Center and the principal investigator for Trestles, said, “During its four-year run, Trestles successfully supported thousands of national researchers with modest-scale applications as well as science gateways, with shorter turnaround times than has been typical for most high-performance computing systems. It’s terrific that this resource can be re-deployed to advance scientific research at the University of Arkansas.”
The collected eXtreme Digital Network facilities, integrated by NSF’s Extreme Science and Engineering Discovery Environment project offer researchers access to a network of supercomputers and high-end visualization and data analysis resources across the country. Trestles continues to operate as part of the network at the U of A.