U.S. leadership in supercomputing technologies is eroding, and with it the broader scientific and technological progress underlying a strong and robust U.S. economy, said Susan Graham at the University of California at Berkeley.

Graham co-chaired the National Research Council's committee to study the future of supercomputing. The committee's report was released this week.

Supercomputing is critical for the United States for at least three reasons, said Horst Simon, associate laboratory director of computing sciences at Lawrence Berkeley National Laboratory in California.

In terms of economic competitiveness, a lot of technology developed and tried out in supercomputers gradually migrates down into the desktop and servers market, Simon told UPI. Therefore, a technological lead in the very high end reflects the state of a nation's computer industry in general.

Supercomputers also are critical for scientific competitiveness, in making sure U.S. scientists lead in fields such as climate science and molecular biology, Simon added.

Supercomputers (also) are very important national security elements, in military applications, management of the stockpile of nuclear weapons for the Department of Energy, and cryptanalysis for the National Security Agency, Simon said.

In the last month, the United States reported great strides in supercomputing. On Nov. 4, the Department of Energy and IBM announced a record-breaking supercomputer performance with BlueGene/L at 70.72 trillion floating-point operations - intensive math calculations - per second, twice as fast as Japan's Earth Simulator, which held the lead for two years at 35.86 teraflops. BlueGene/L is designed to help assess aging nuclear weapons.

In addition, in October NASA announced its most powerful supercomputer, Columbia, also beat the Earth's Simulator with a performance of 42.7 teraflops.

The new systems are one sign that supercomputing continues to be strong in the U.S., Simon said.

Still, whether the most powerful and most expensive supercomputer is located in the United States or elsewhere does not indicate loss or gain of leadership in supercomputing technology, said Marc Snir of the University of Illinois in Urbana-Champaign, co-chair of the National Research Council's supercomputing committee co-chair.

Rather, our concern is that current investments and plans are not sufficient to provide the capabilities that our country needs, Snir told UPI.

Simon said the Earth Simulator's debut in 2002 was a real wakeup call for the U.S. supercomputing establishment, since it was at least four times faster than any machine then in existence.

It was so far ahead of comparable systems in the U.S. that it could get results theoretically impossible to obtain on American supercomputers at that time, he recalled.

Supercomputing in the United States has fallen behind, Simon said, because the American way of approaching the field, if there was said to be a supercomputing policy at all, was to leverage commercial off-the-shelf technology.

Many current leading supercomputers are clusters of personal computers linked to work together. Making supercomputers out of clusters was exactly the right approach at the time it came out, to leverage the microprocessor revolution, but by 2002, I would say that approach was seen as having outlived its usefulness, Simon said.

For the past decade, insufficient government funding for supercomputing, little long-term planning and inadequate coordination among relevant federal agencies also have hampered opportunities to make the most of supercomputing, Graham said.

The government should make long-term plans for sustained investments into supercomputing research and development rather than gunning for whether the United States has the fastest supercomputer in the world, Graham's committee recommended.

Such plans would ensure stability in the marketplace, enabling supercomputing hardware and software vendors to remain viable and grow.

This stability is critical because many of the machines needed for the most demanding supercomputing applications - particularly for national security - must be custom-built, the committee added.

The committee noted the U.S. government also should support the creation and maintenance of supercomputing software.

It's three or four times faster to build new hardware than it is to build a complicated, large software environment currently used in high-end computing, Simon said.

"That means the hardware cannot change too radically. A faster novel system that is radically different may be good for a few applications, but not for software development in general.

Hardware and software need to be co-evolved, which was not really practiced in past, he explained. Vendors then came up with fantastic new supercomputers and hoped the software and applications would fall into place.

To meet these goals, the committee said government agencies should develop a single, integrated plan to allow them to identify common needs at an early stage, leverage resources and minimize duplicative efforts. The agencies and Congress would then use this plan to coordinate annual investments and track progress. This plan would replace the current diffuse interagency structure.

An integrated plan is not an integrated budget, the committee noted. The use of such a plan would not preclude agencies from individual activities, nor would it prevent them from setting their own priorities - while long-term basic research matches the National Science Foundation's mission, industrial supercomputing research and development is more akin to DARPA's mission.

Steady investments over time and clear milestones for the future should provide returns on investment that greatly exceed the original costs, Graham said.

Simon said BlueGene/L could represent an important new direction for supercomputing. BlueGene/L, when complete, will have more than 65,000 computing nodes, compared with the Earth Simulator's 5,120.

Though it will have more than 12 times as many processors, the finished BlueGene/L actually will require one-seventh as much electrical power and one-fourteenth the floor space as the Earth Simulator because BlueGene/L's chips are not as powerful. This means they require less power and cooling, and can be packed closer together.

BlueGene/L may be the first example of many more systems that come with not thousands but tens of thousands of processors, Simon said.

As a consequence, the software will need to adjust as well. Instead of divvying a problem up among 5,000 processors, software designers will have to find a way to split problems across more than 60,000, and not all applications will be fit to run at such a chopped-up level.

All rights reserved. Copyright 2004 by United Press International. Sections of the information displayed on this page (dispatches, photographs, logos) are protected by intellectual property rights owned by United Press International. As a consequence, you may not copy, reproduce, modify, transmit, publish, display or in any way commercially exploit any of the content of this section without the prior written consent of by United Press International.

Related Links
SpaceDaily
Search SpaceDaily
Subscribe To SpaceDaily Express

Thanks for being here; We need your help. The SpaceDaily news network continues to grow but revenues have never been harder to maintain. With the rise of Ad Blockers, and Facebook - our traditional revenue sources via quality network advertising continues to decline. And unlike so many other news sites, we don't have a paywall - with those annoying usernames and passwords. Our news coverage takes time and effort to publish 365 days a year. If you find our news sites informative and useful then please consider becoming a regular supporter or for now make a one off contribution.
SpaceDaily Contributor $5 Billed Once credit card or paypal		SpaceDaily Monthly Supporter $5 Billed Monthly paypal only