Strength in Numbers–High-Performance Computing

For most of us, one computer suffices. We’ve got it all— operating system, processor, memory, and we’re up and running. For some, however, only multiple computers will do, and that’s where high-performance computing answers the call.

Simply put, high-performance computing involves a collection of computers that work together to solve a problem. This is known as a computer cluster, which might be thought of as similar to a server. UMass Boston has two such clusters on campus. One cluster has nine computers, referred to as nodes; the other, 36. A third cluster, located at the Massachusetts Green High Performance Computer Center in Holyoke, Massachusetts, offers researchers 13,000 processors.

A cluster can handle far more complex data operations in far less time. Jeff Dusenberry, director of Research Computing, says IT staff members maintain the clusters and are available to troubleshoot any issues that come up while researchers conduct their work.

High-performance computing is available to faculty in diverse disciplines. “Not every job is suitable for high-performance computing,” says Dusenberry. “It has to be a job that can break down into smaller pieces and do that efficiently. And different topics are handled differently. In some biology analyses, for example, you might tell each computer to do the same thing but with different data.”

In the end, based on the parameters input by researchers, the cluster will bring the data back together and calculate the results.

Chemistry professor Jason Green supervises a team of researchers, including two postdoctoral researchers, three graduate High-Performance Computing students, and two undergraduate students, “all trained in some way or another by Jeff,” says Green.

In Green’s case, high-performance computing is used to answer very carefully crafted questions about chemical systems. “By simulating models of these systems, we test hypotheses, refine our models, and make connections between theory and experiment in order to develop high-fidelity predictions,” he says. “Model building and simulation are essential to determine which details of a chemical system are important and which can be neglected.”

Sound complex? Think of it this way: one person, working with one computer, is limited by the speed at which their computer can process data. For certain problems, calculations could take years or decades on a single processor. By running calculations in parallel across many different processors, that time can be reduced to a few hours or weeks.

Green and his team of researchers work with a computer cluster and ask intensely complex scientific questions. “We do projects where calculations can span months, even years, on our computer clusters,” says Green. “These calculations are way beyond the capabilities of a single computer.”

The bottom line here is that high-performance computing can save days, weeks, or more of research time by parsing out different elements and components of the project to different nodes in the cluster. Throughout the process, Dusenberry and his team are available to lend a hand when a hand is needed.

“Jeff and his team have spent countless hours troubleshooting hardware and software issues, training students, and maintaining the clusters,” says Green. “Jeff is extremely important to the success of our research.