from February 4, 2002 Knoxville
Shared resources will help address super software needs
By Frank Munger News-Sentinel, senior writer
February 4, 2002
Oak Ridge National Laboratory is one of the nation's premier supercomputing centers,
boasting a new IBM machine (nicknamed The Cheetah) that soon will be capable of 4
trillion calculations per second, a level known in the computer world as 4 teraflops.
But when ORNL and other research institutions receive these rare or, in some instances,
one-of-a-kind supercomputers, they don't receive all the software necessary to run them
Typically, the laboratories come up with their own "homegrown" software to deal with such
things as scheduling research projects on the computers and monitoring their operations.
"Those things are usually site-specific," said Al Geist, who works on ORNL's advanced
Providing tailored software is really not a priority for the computer manufacturers, who
make their business living on mid-range systems, Web servers and database farms, he said.
"They're probably only making five or six of these (supercomputers) per year, and they're
selling them to people who expect to get the software for free," Geist said.
So the supercomputer centers are left to fend for themselves on time-consuming and
increasingly complex software needs.
Now, however, ORNL and a team of other national labs and universities have pooled their
minds and money for a five-year, $15 million project called the Scalable Systems Software
Center. Together they will share ideas and address mutual problems in managing the bigger
The effort was launched last fall, and progress is already apparent, Geist said.
Besides ORNL, other national labs participating are Argonne, Ames, Lawrence Berkeley,
Los Alamos, Pacific Northwest and Sandia. Also, the National Center for Supercomputer
Applications, representing dozens of universities, is involved.
Geist said the computer manufacturers, such as IBM and Compaq, also have been attending
the meetings. "Even Intel has been represented," he said. "They're interested because of
their involvement in PC clusters."
The expectation is that the collected expertise will come up with some standardized
software solutions, such as the interfaces between system components.
Different institutions have different specialties, and so various tasks are being
farmed out, such as development of monitoring software. A problem solved at one
supercomputer center can be leveraged to help others facing similar difficulties.
Not only will the effort make the high-performance supercomputers more functional,
but it should save money as well, Geist said.
He noted: "As these machines get bigger and bigger (10,000 processors, in
many cases), you don't want to have to increase the number of people it takes to run them.
If it takes 10 people to run a one-teraflop machine, you don't want to have to hire 100
when you go to 10 teraflops."
Copyright 2002, KnoxNews. All Rights Reserved.
Mirrored with permission