Update: Intel, IBM, Qwest Power "TeraGrid"
ASCI White to Be Formally Unveiled Friday
August 9, 2001 By: Mark Hachman

¡¡Intel Corp. today announced that more than 3,300 Itanium processors will eventually be used to build a distributed, scientific computing system expected to be the largest of its kind in the world.
¡¡The computing system, dubbed the "TeraGrid," is part of a $53 million award by the National Science Foundation (NSF) to four facilities to address complex scientific research by creating a Distributed Terascale Facility (DTF with approximately 13.6 teraflops of computing power.
¡¡Since the computing clusters will be deployed during the third quarter of 2002, all of the Itanium 64-bit processors will actually be from the second-generation "McKinley" family, according to Dick Bland, global manager of high-performance computing at Intel. The chips will be off-the-shelf parts, not optimized in any way.
¡¡IBM, Armonk, N.Y., will deploy the processors inside roughly 1,000 Linux-powered of its eServers, together with about 600 terabytes of data storage. The four sites will be linked by a 40-Gbit/s fibre-optic network, four times faster than the 10-Gbit/s network today, supplied by Qwest.
¡¡The concept of distributed computing ¨C computing nodes shared across meters or kilometers of physical distance, rather than physically next to each other in a shared multiprocessor (SMP) system ¨C is fairly evolved, although the first practical implementations have only been deployed in the past two years or so.
¡¡Theoretically, the scalability of distributed computing is almost limitless, assuming that the applications running on top of it are tolerant of any latencies the physical distances cause, according to Dan Powers, director of early-stage Internet technology at IBM.
¡¡"There are several practical limitations--money is one of them," pointed out Dan Reed, director of the National Center for SuperComputing Applications. "As you build clusters larger and larger, power and cooling also become an issue. But IBM's right; there are not intellectual challenges to building, say, a 50 Tflop cluster."
¡¡The other advantage of distributed computing is its ability to allow access to scientific instruments to the industry at large. By itself, the R/S6000 based "ASCI White" supercomputer developed last year is capable of 12.3 Tflops by itself. But officials pointed out that it was designed for a specific application; with the TeraGrid, scientists can not only run jobs on the network at large, but access shared instruments, such as a radio telescope or other device.
¡¡ASCI White is scheduled to be officially unveiled Friday morning at the Sandia Labs in Livermore, Calif.
"What this means is that in the past a guy got two hours on the machines in the middile of the night," Intel's Bland said. "Now that we've got industry standard building blocks, I can get this in my lab¡­have the code up and running on a smaller cluster, and be ready to go."
¡¡For the past year, IBM and others have also tried to promote the use of "on-demand" computing and storage on an as-needed basis. Grid computing takes this a step further, sharing computing in a manner similar to jobs that are run on single-location mainframes. The concept also leverages the peer-to-peer concepts that is making the KaZaa network a success. The academic environment is an ideal test bed for these concepts, officials said, as they can figure out which approaches will work best before asking Fortune 500 customers to run mission-critical enterprise applications upon them.
¡¡"This is the beginning of a national cyber-infrastructure," said Fran Berman, the director of the San Diego Computer Center, one of the facilities involved. Chicago's Argonne Laboratory and the California Institute of Technology in Pasadena will also host the TeraGrid.
¡¡NCSA currently has a 2-Tflop cluster, one made up of IA-32 processors and one an Itanium cluster. Of the 11.6 Tflops of total computing power the award will fund, a total of 6.1 Tflops will be at the center itself.
¡¡In addition to providing the processors powering the IBM systems, Intel will supply the TeraGrid with key compilers, software, tools and engineering design, and tuning support services.