from February 24, 2003, News-Sentinel

original URL:
http://www.knoxnews.com/kns/news_columnists/article/0,1406,KNS_359_1766645,00.html

Budget slowspace of faster ORNL computing

By Frank Munger, munger@knews.com
February 24, 2003

Budget slowspace of fasterORNL computingOak Ridge National Laboratory's plan to have the world's fastest supercomputer in 2004 suffered a blow when the Bush administration's proposed federal budget for that year was released recently.

The amount of funding for next-generation computer architectures was proposed at $14 million, much less than the $100 million or so needed to maximize research opportunities and challenge the Japanese for worldwide computing superiority.

"That (goal) is no longer feasible, based on the president's budget,'' Thomas Zacharia, the lab's computing chief, conceded. Zacharia hopes the setback will be temporary, and he said ORNL's budding relationship with Cray Corp. is still intact and ready to blossom.

The first Cray machine should arrive within the next month, providing Oak Ridge researchers with the top U.S. supercomputer featuring vector-type architecture.

The initial Cray X-1 machine will operate at "only'' 3 or 4 teraflops (a teraflop represents a trillion calculations per second), which is similar to the speed of the IBM Cheetah at ORNL. But the Cray X-1 and the next-generation Cray X-2 (code-named Black Widow) offer incredible scale-up opportunities.

The computing potential is somewhere in the range of 800 teraflops, according to Zacharia.

Still, Oak Ridge had hoped to pass Japan's Earth Simulator (40 teraflops) within the next two years, and now officials are forced to re-evaluate those timetables.

"Even if we had all the funding, we still could not go out and buy a 40-teraflop machine right away,'' Zacharia said. "If the '05 budget recognizes the merits of the program, the delay could be as little as six months to a year.''

The Oak Ridge lab is gearing up to make a strong case for a big funding increase in fiscal 2005.

The scale-up in computing capabilities is hugely important for the U.S. science enterprise, with a near-term need for 50 to 100 teraflops of computing power, Zacharia said.

"The research applications range from nanoscience to biology to astrophysics to climate to fusion,'' he said. "All of these things are very important to Oak Ridge.''

Zacharia obviously is disappointed that aggressive goals set by ORNL won't be met, but he said he is encouraged by activities under way at the laboratory.

"We have the people, we have the programs, we have the facilities, we have the networking and the partnerships, and we have a very solid technical plan,'' he said. "Having said that, clearly our deployment could get delayed. The question is how much. That depends on what happens in '05.''

ORNL may not meet its goal of having a petaflop machine (capable of 1,000 trillion calculations per second) in 2007, but progress is being made, Zacharia said.

Funding for advanced computing at ORNL is about $25 million this year, roughly the same as in 2002, and the 2004 funding looks likely to stay in that range again, he said.

"At least the president's budget will allow us to begin the process,'' the lab official said. "At the time we put this plan together, of course, we were not talking about a war in Iraq and other budget pressures. Given all those pressures, we are doing very well.''

Senior writer Frank Munger may be reached at 865-342-6329 or munger@knews.com.