Teraflops and Atom Smashes

The partnership of UT and Oak Ridge National Lab promises a mind-bending array of resources for both

Given the densely striking configuration of the computer room upstairs at the Oak Ridge National Laboratory's central compound, it's no stretch for a visitor to presume any number of visually analogous locales: a rigidly staged appliance showroom; a geometrically impenetrable Star Trekkian engine vault; an elaborate garden maze, buttressed by towering black consoles rather than neatly tailored hedges...

But those consoles, grave and dark and covered with hundreds of tiny ventilation eyelets, like a rock concert's phalanx of thundering Marshall speaker cabinets, are actually the seats of nigh-inconceivable computing power. Each of the 13 cabinets is home to some 16 10' x 8' components, each component comprising a computing unit unto itself, each with its own power supply and processors and network connections. With four processors per computer and 64 per cabinet, the entire system with its more than 700 total processors is capable of more than a trillion calculations per second. (Compare that number with perhaps a half-billion, the outer limit for the average 500 megahertz PC.)

Such are the vital statistics of the laboratory's IBM RS/6000 SP, the 1.5 teraflop supercomputer that constituted the world's largest and most powerful non-classified computing system upon its purchase just a few years ago. Dr. Thomas Zachariah, deputy assistant director for the laboratory's high performance computing division, explains that "teraflop" is techno-jargon for a trillion floating point operations per second.

"Take an ordinary calculator, and punch in a number the size of all the numbers on the keyboard," he says, reiterating the concept in layman's shorthand. "Then multiply that number by a number the same size. This computer can do a trillion such calculations in a second, 60 trillion calculations in a minute."

Such overwhelming parallel processing capabilities allow ORNL scientists to surmount any number of preposterous research obstacles: to reassemble millions of sundered DNA components into an entire sequence; to determine the properties of materials by way of the very atoms that comprise them; to construct long-term global climate models that take into account hundreds of atmospheric variables across hundreds of thousands of individual geographic sectors.

Beginning roughly July of 2001, the RS/6000 will also allow University of Tennessee student and faculty researchers to breach the outer limits of high performance computing, serving as the linchpin of the impending UT/ORNL Joint Institute for Computational Sciences.

The year 1999 saw UT and its ally Battelle Institute make a bid for the role of prime contractor at the Department of Energy's Oak Ridge laboratory, a bid cast in the wake of the expiration of former overseer Lockheed-Martin's DOE contract. February of 2000 saw UT-Battelle win the contract and subsequently announce the launching of four so-called "joint institutes," lab-university collaborations (with accompanying new physical facilities) intended to avail UT researchers of the lab's abundant resources and augment the school's research profile.

Those collaboratives—which include the JICS, the Joint Institute of Neutron Science, the Joint Institute of Biological Sciences, and a think tank called the Oak Ridge Center for Advanced Studies—will be nurtured at the outset by state funding, to the level of some $26 million total over the next four to five years. And while all four institutes offer boundless prospects for groundbreaking study, it is the JICS that has perhaps the most astounding research potential, and which will serve as an early barometer of the efforts yet to come.

According to UT director of communications and community outreach Billy Stair, the influx of $6 million will enable the construction of the new JICS facility on the main campus of ORNL, as well as the augmentation of the supercomputer's capabilities from a "mere" 1.5 teraflops to 10 teraflops. The partnership has thus been a fruitful one for both institutions, as the state funding was contingent on the success of UT's DOE bid.

"UT would not likely ever be able to house something of this magnitude," says Stair. "And without UT, ORNL wouldn't have had access to this kind of extra state funding."

Dr. Lee Riedinger, deputy director of science and technology at ORNL, says the joint institutes will serve as the loci for an evolving cast of researchers and projects—projects drawing on one of the institutes' three key resources (the JICS supercomputer; the ORNL Mouse House, an ongoing lab initiative to unravel genetic mysteries through study of generations of mice bred and cataloged since World War II (JIBS); and the pending Spallation Neutron Source, the billion-dollar neutron scattering device that will revolutionize that branch of physics and provide the fulcrum of the JINS).

"I see these institutes as being lean on staff," says Riedinger. "After the state funds the physical facilities, we'll lobby DOE and other sources for operational funds. Then the emphasis will be more on visitors and workshops than permanent staff."

As stated by another UT researcher, the joint institutes will constitute the "bricks and mortar infrastructure" for any number of interrelated initiatives.

Early funding for the JICS, which will eventually coexist with the Oak Ridge think tank in a single facility, has been earmarked for the state's 2002 budget, an arrangement that will actually allow the project to launch this July. According to Stair, the funding for both the JICS ($6 million) and Oak Ridge Centers ($4 million) will be fully allocated by 2004, at which point the JICS supercomputer should also have reached its 10 teraflop zenith.

"All of this serves as a very clear recognition of the importance of high performance computing in the lab's mission," says Zachariah. "High performance computing is an incredibly important scientific enabler, rapidly evolving, a key to training the next generation of researchers."

Within the cloistered geometry of the computer room, ORNL's Buddy Bland explains how the laboratory's supercomputing capabilities have already been deployed in figuring scientific conundrums of unfathomable complexity, and how the impending trillions-per-second increase in computational efficiency will allow researchers to take exponentially more precise measure of those problems.

Says Bland, computers such as the RS/6000 were vital in sequencing the human genome, the recent genetics breakthrough that promises a quantum leap in medicine and diagnostics over the coming years.

"To sequence DNA, you divide a strand into pieces, then run those pieces through machines to identify components of the chain," says Bland. "But each sequence is hundreds of thousands of components long, and the original strand is even longer. You have to paste those components back together, like a puzzle with a million pieces."

Via supercomputing, the sequencing can be spread across a number of high-efficiency processors, enabling like strains to be identified, matched and reassembled at enormous speeds. The next frontier in genetics research will be functional genomics—identifying the role each genome component plays in various biological processes—an endeavor to which Bland says supercomputing will be "indispensable."

In an equally ambitious undertaking, the RS/6000 has been utilized in a national exercise in global climate modeling, an effort spurred by the 1998 Kyoto international environmental conference. The goal, says Bland, is to construct a comprehensive model of the earth's climate that extends more than 100 years into the future. The variables relevant to that model—ocean currents, temperatures, rainfall, ad infinitum—call for innumerable separate individual models, each of those extrapolated for many divergent possibilities.

What's more, says Bland, the goal is to sharpen the focus of that climatic blueprint even further. All of those possible variables and iterations must be figured for individual geographic sectors—square "gridpoints" of roughly 200 x 200 kilometers. And increasing the already considerable resolution of the model will only be possible with ever more powerful supercomputers, an ever-larger number of processors over which to spread those calculations and components.

"What we'd really like to do is go to even finer resolution, maybe 20 x 20 kilometers," says Bland. "That's an enormous number of gridpoints, a hundred-fold increase from the number we have now."

Given the epic scope of these current research efforts, and given resources of the magnitude promised by the realization of the joint institutes' funding program, UT/ORNL administrators believe the lab and university together will be unique in their profile, in their ability to lure top minds in certain areas of scientific endeavor; a "focal point" for top scientific researchers, says Zachariah.

Concurs the University's Dr. Frank Harris, director-elect of the JIBS, "The joint institutes can be magnets for top students, top researchers, even for local economic growth. But it's just like reassembling a DNA sequence. We have to fit the pieces together, figure out how to make it work."

© 2001 MetroPulse. All rights reserved. This material may not be published, broadcast, rewritten or redistributed.

Comments » 0

Be the first to post a comment!

Share your thoughts

Comments are the sole responsibility of the person posting them. You agree not to post comments that are off topic, defamatory, obscene, abusive, threatening or an invasion of privacy. Violators may be banned. Click here for our full user agreement.

Comments can be shared on Facebook and Yahoo!. Add both options by connecting your profiles.