Planning, placement and more: Optimization makes it easier

(page 3 of 3)

“Parallel computers allow you to solve problems that can’t be solved” by other means, she adds.  “If there’s a problem where the difference between an approximation and an optimal solution has serious consequences for the cost to the government, or for national security, if you’re willing to put in the time and the effort to run this on a big machine, that’s what our (computer) code is for.”

The techniques Phillips and her fellow researchers have developed can even improve the performance of parallel computers themselves.

Large parallel machines with thousands of processors often have several jobs running at one time, Phillips says.  “Each job might be allocated to some subset of processors, and it could be that having those jobs close to each other is important in terms of not having their communication interfere with others, and having fast response to their own communication,” she adds.

Phillips’ group created the Compute Processor Allocator (CPA) to parcel out processor nodes based on location and to balance job allocation with future allocations on large computers.  In experiments, CPA increased locality and processor production on a parallel computer by 23 percent over other allocators.

The program earned an R&D 100 Award as one of the top new technologies of 2006.

Now Phillips and her fellow researchers are working with computer scientist Ali Pinar of the DOE’s Lawrence Berkeley National Laboratory on the “Supernova Factory Scheduler,” a program designed to maximize telescope time to find and monitor exploding stars.

Other new applications include tracking particle movements in supercolliding “atom-smashers” and detecting groundwater contamination.

In fact, Phillips says, there are so many uses for optimization codes that “I sometimes refer to our group as mathematical mercenaries, since we can usually find applications to work on” regardless of circumstances.

« Previous       1   |   2   |   3   |   Print