SuperCluster Makes Computation Speeds 20 Times Faster

Computerworld - This version of the story originally appeared in Computerworld's print edition.

To get a feel for the massive processing power Northrop Grumman garnered through its recent high-performance computing initiative, consider the fact that it would take a single workstation nearly three years to process the amount of data that the new "SuperCluster" can now churn through in half a day.

The aerospace giant needs sky-high processing power so that scientists and engineers working on spacecraft design and other projects can perform complex computations on the massive amounts of data pulled from an array of satellites orbiting the planet.

For instance, a single climate-sensing satellite might carry 45,000 sensor elements — each extracting thousands of images. Northrop Grumman engineers must use all that data in complex modeling, simulation and analysis exercises.

Best in Class

This story is part of an ongoing series showcasing the best projects of this year's Premier 100 IT Leaders.

Northrop Grumman Space Technology Sector

This $3.1 billion unit of Los Angeles-based Northrop Grumman Corp. develops electronics for military and space systems.

IT Champion: Brad Furukawa, vice president and CIO

IT Staff: 550 people

IT Budget: $150 million

Project Payback: Increased processing time by a factor of 20; integrated 400 complex engineering applications into a single infrastructure; retired more than 300 Unix workstations; improved overall efficiency and reliability of engineering design.

To meet the processing demands of its engineering and scientific community, Northrop Grumman's IT staff installed a cluster of 1,800 Linux-based CPUs that are capable of performing parallel processing. The new SuperCluster high-performance computing system is 20 times faster than previous systems.

By year's end, it will have 6,000 CPUs.

"Prior to the availability of our SuperCluster, individual departments and programs had developed high-end computational solutions, such as Beowulf clustering, turnkey multiprocessors and compute farms," explains Brad Furukawa, vice president and CIO of Northrop Grumman's Space Technology Sector in Redondo Beach, Calif. "Maintaining these small stand-alone clusters proved to be highly inefficient."

Improving efficiency took more than additional CPUs, he says, emphasizing the need for better load balancing and increased storage capacity, which is now 20TB collectively.

"Our high-performance-computing SuperCluster is more than just a large number of processors," says Furukawa. "Given the configuration of job types in use, the local storage is a critical part of serving large amounts of data to the many processors working simultaneously."

Northrop Grumman's move away from scattered computational resources reflects a trend among organizations that need massive processing power.

"Many of our clients are building similar computing resources, where previously they had many independent islands of computing in the form of stand-alone clusters or scale-up systems," says Carl Claunch, an analyst at Gartner Inc.