While the new concept of grid computing has started to flourish in the academic community, officials with IBM Corp. already are eyeing corporate and government applications that coul
|  |
d follow recent experimental projects the company has won.
On Aug. 9, along with Qwest Communications International Inc., Denver, IBM was chosen to support t
| |
he National Science Foundation’s three-year, $53 million grid computing project. This news came shortly after the company announced it was developing two grids abroad: one for the British government’s e-Science Core Programme, a $72 million package to develop a global scientific collaboration infrastructure; and a second one for the Netherlands.
David Turek, IBM vice president of emerging technologies, said IBM of Armonk, N.Y., will build $4 billion worth of grid-related projects over the next few years.
“It especially has attracted a lot of interest by large companies that can use it to connect geographically diverse offices and unify the supply chain,” he said.
It can also be used to create virtual databases that pool the content from different offices, and take on large engineering and modeling projects beyond the scope of any single computer.
Computational grids aggregate computers to pool data, applications and processing power, said Richard Hirsh, a National Science Foundation program officer of the Distributed Terascale Facility project. For some time now, servers have been networked in clusters close to each other to offer virtual-supercomputer levels of performance.
Grid computing goes to the next step: It links computers over geographically remote areas to perform as a single, virtual supercomputer.
The National Science Foundation’s network will link four high-performance computers into a single, virtual entity, allowing researchers to tackle problems in weather modeling, protein folding, nuclear modeling and other areas previously beyond today’s computational powers.
This project is a joint undertaking of the San Diego Supercomputing Center, the California Institute of Technology in Los Angeles, the National Center for Supercomputing Applications (NCSA) in Champaign-Urbana, Ill., and the Argonne National Laboratory in Chicago.
“Anyone can build a smaller scale version of a cluster like this,” said Hirsh, pointing out that the NCSA already offers free basic software called “Grid-in-a-Box” to build demonstration grids.
IBM’s role in the National Science Foundation grid is primarily one of systems integrator, Hirsh said. IBM will provide clustered Linux servers based on next-generation 64-bit Intel Itanium family processors. IBM supercomputing software will handle cluster and file management tasks.
The grid networking protocols and application program interface tools were developed by the open source Globus Project, an effort by the Department of Energy, the University of Southern California and independent developers.
IBM also will use cluster-scalable, fiber interconnects from Myricom Inc., Arcadia, Calif., switching equipment from Cisco Systems Inc., San Jose, Calif., database solutions from Oracle Corp., Redwood Shores, Calif., and its own TotalStorage products. Qwest will provide a dedicated 40 gigabit optical network to connect the locations.
The facility will go online by mid-2002 and be fully operational by April 2003. According to George Strawn, the executive officer for National Science Foundation’s Directorate for Computer & Information Science and Engineering, the grid will be stable enough for everyday use by scientists. But as researchers are drawing on this resource, it will also allow developers to work out the bugs of the design.
