Atmospheric research center plugs into grid computing
Connecting state and local government leaders
The National Center for Atmospheric Research is ramping up for grid computing with a test bed for managing multiple supercomputing resources under a single interface.
The National Center for Atmospheric Research is ramping up for grid computing with a test bed for managing multiple supercomputing resources under a single interface.
The National Science Foundation-funded center in Boulder, Colo., chose Platform LSF HPC from Platform Computing Inc. of Toronto to manage workloads on a test bed with a 256-processor Linux cluster and a 64-processor IBM eServer p690 AIX cluster, plus some smaller machines. The price was not disclosed.
'We want to take advantage of commodity clusters and familiarize ourselves with grid computing,' said high-performance computing specialist Tom Engel. 'Over the years we've had all kinds of supercomputers on the floor from Cray, IBM, SGI and others. Each had its own homegrown subsystem to control workloads.'
Now, he said, 'We can begin standardizing the computing environment with the single interface' of LSF. 'Users won't have to reconstruct their job scripts' and learn a new operating system, regardless of which vendor NCAR chooses for its next supercomputer.
NCAR's staff of 1,300 scientists from 68 partner universities work on world climate and weather models, via NSFnet and the Abilene research network. The test bed users currently have a one-time password token for Secure Shell access to an NCAR front-end computer. Secure Shell logs them in over a network and executes their program commands remotely.
'Eventually we will have a Web portal interface to the LSF scheduler,' which lets center officials set a requested job's priority and charge for measured use, Engel said.
The center has total computing capacity of more than 12 trillion floating-point operations per second. Its IBM Blue Sky cluster could become 'the biggest p690 cluster in the world' in early 2006, he said. 'We hope to acquire a single-frame IBM Blue Gene next spring.'
NCAR hopes eventually to join NSF's distributed TeraGrid, which last month went into full production mode at nine supercomputing sites.