Researcher To Make Supercomputers Compute Even Faster

COLUMBUS , Ohio – A prestigious award from the National Science Foundation will help Umit Catalyurek design new tools to speed up the analysis of massive data sets that can be untangled – and therefore understood – only by using extremely powerful computers and supercomputers.

The assistant professor of biomedical informatics and electrical and computer engineering at Ohio State University has earned one of NSF's Faculty Early Career Development (CAREER) awards, which recognize a young researcher's dual commitment to scholarship and education. He will receive $400,000 in funding over five years for an ambitious project titled “Scalable combinatorial scientific computing.”

Catalyurek will use the award to design efficient algorithms to enable increasingly large parallel computing systems to solve ever more complex scientific problems. An algorithm is a mathematical formula with step-by-step procedures to solve a given problem. A parallel computing system is a system in which thousands to hundreds of thousands of computer processors work cooperatively to solve a single problem.

“Current algorithms are designed to run on systems with up to 1,000 processors, maybe,” Catalyurek said. “But we now have machines and networks that have 10,000, even 100,000, processors.

“While these machines hold great promise for scientific applications, the problem is that we don't have the right kind of algorithms to tell these machines how to handle data,” he continued.

“The scientific community is handling increasingly complex and computationally challenging problems,” Catalyurek said. “The use of combinatorial algorithms will let many fields of science use large-scale computing where it just wasn't possible to do so before.”

Scientists use algorithms and large-scale computing to interpret data from a host of different disciplines, including data-heavy images from digitized microscopes or MRI machines, and even figuring out where are the best areas to drill for oil. While supercomputing allows researchers to model and interpret data collected from such sources, current supercomputing methods typically force their users to settle for approximations of what the data really represent.

“It's very difficult to create a true-to-life model with so much data,” said Catalyurek.

He will use the CAREER award to develop a special set of algorithms, what scientists call combinatorial algorithms. These instruction sets provide many different options for analyzing specific problems. The beauty of combinatorial algorithms, says Catalyurek, is that such instructions can allow for more accurate results as well as the faster execution of those results.

“The scientific community is handling increasingly complex and computationally challenging problems,” Catalyurek said. “The use of combinatorial algorithms will let many fields of science use large-scale computing where it just wasn't possible to do so before.”

Catalyurek's work will help researchers create clearer models of their data using hypergraphs, graphs that show multi-way interactions between multiple objects.

“Instead of looking at the interaction between two objects, we can get a clear idea of the interactions between multiple objects,” Catalyurek said. “That ultimately gives users a better idea of what their data actually represent.”

Catalyurek is Ohio State's 55th CAREER award recipient since 1996.

#

Contact: Umit Catalyurek; (614) 292-0914; Catalyurek.1@osu.edu

Written by Holly Wagner; (614) 292-8310; Wagner.235@osu.edu