Blue Waters: Changing the way science is done

released 08.11.08

NCSA's Rob Pennington and Thom Dunning
Rob Pennington, NCSA Deputy Director and Thom Dunning, NCSA Director

University of Illinois, NCSA, IBM, and partners to build first sustained-petascale supercomputer

Extending more than 50 years of supercomputing leadership, the University of Illinois at Urbana-Champaign and NCSA will be home to the world's first sustained-petascale computational system dedicated to open scientific research. To be called Blue Waters, this computer is supported by the National Science Foundation and will come online in 2011. Recognizing the promise of petascale computing, the state of Illinois and the University of Illinois have pledged additional support so that researchers can fully exploit Blue Waters' unique capabilities. This support will include construction of a new facility to house the system.

Blue Waters will sustain performance of at least one petaflop on many applications that are crucial to researchers—not on benchmarks, on real-world codes scientists and engineers use every day to simulate things like the weather, the stars, and new medicines. That's more than 1 quadrillion calculations per second.

In addition to raw computing power, the Blue Waters project also will include support for application development, system software enhancements, interactions with business and industry, and educational programs. This comprehensive approach will ensure that users across the country will be able to use Blue Waters to its fullest potential.

To create a broad range of achievements, Blue Waters will require an intense, years-long collaboration among Illinois, NCSA, IBM, and the dozens of universities, colleges, research laboratories, and institutes that have formed the Great Lakes Consortium for Petascale Computation. This partnership is dedicated to encouraging the widespread and effective use of petascale computing to advance scientific discovery and the state-of-the-art in engineering, to increasing regional and national competitiveness, and to training tomorrow's computational researchers and educators.

Principal investigators on the Blue Waters project are: NCSA's Thom Dunning and Rob Pennington, the University of Illinois' Wen-mei Hwu and Marc Snir, and Louisiana State University's Ed Seidel.

Thom Dunning, Director, NCSA
Rob Pennington, Deputy Director, NCSA

Until about the middle of the last century, science was really founded on two major premises. One was work in the laboratory, referred to as experiment; the other was work on discovering the underlying principles, theory. With the development of electronic computers in the 1950s, scientists began to realize there was actually a third mode of investigating the world around us: computational modeling.

Computational modeling means taking the mathematical equations that describe the phenomena we're interested in and then using the computer to solve those mathematical equations because there is no other way to solve them. But as models became more accurate they became more sophisticated and more computationally intense. To address this problem, some University of Illinois professors sent an unsolicited proposal to the National Science Foundation (NSF) regarding the need for supercomputer power; NSF responded in 1985 by establishing five supercomputer centers, of which NCSA was one.

As the world has changed in the last two decades, so has science and engineering. The fidelity of the models researchers use is increasing, which means they’re able to predict what is going to happen in the real world with greater accuracy and greater precision. But this requires yet more computing resources. Blue Waters, the sustained petascale computing system to be built and deployed in 2011 by the University of Illinois, its National Center for Supercomputing Applications, IBM, and our partners in the Great Lakes Consortium for Petascale Computation, will provide those resources.

Through the years we've learned that scientists and engineers are more interested in the sustained performance of a system, the performance level that the machine will achieve when running their codes, than the peak performance number, which tells the theoretical maximum performance of the machine. With Blue Waters, scientists and engineers will have sustained petascale performance—the capability to process one quadrillion calculations per second—affording them additional accuracy to describe the phenomena in which they're interested. Thus Blue Waters is going to make it possible to do science on scales not seen before as researchers solve problems that couldn’t be solved previously.

What type of problems?

Blue Waters ultimately is going to lead to far more accurate predictions in severe weather events such as tornadoes and hurricanes. Other uses for Blue Waters could be designing new materials with specific properties, plastics for example. Pharmaceutical design is another use, as are better predictions with many other aspects of health and medicine. Astrophysics questions like what happens inside a star at the end of its life may be answered. We also expect Blue Waters to be used by several of our industrial partners to tackle problems they consider to be critical to their competitiveness in this age of globalization, such as designing more lightweight, safer airplanes or improving communication technologies.

Blue Waters will be the type of system that researchers go to when they have a problem that doesn't fit anywhere else. It will be the system for problems that require solutions in a very short period of time, or can't ever be solved on a researcher's computing system in the lab or even on a regional computing environment.

Blue Waters is going to be one of the leading research instruments of the coming decade, opening ever more phenomena to researchers' interest and exploration.

del.icio.us

Slashdot