Astrophysicists and cosmologists at the University of California, Santa Cruz, are among the first scientists to have access to the powerful new Columbia supercomputer at the NASA Ames Research Center. The UCSC scientists have been using the new system's unprecedented computing power to run simulations of complex phenomena such as supernova explosions, gamma-ray bursts, and dark matter halos.
NASA will be featuring the work of UCSC researchers along with various NASA projects in demonstrations and presentations this week at SC2004, the premier international conference on high-performance computing, networking, and storage. The conference takes place at the David L. Lawrence Convention Center in Pittsburgh, November 6-12.
The Columbia supercomputer, named to honor the crew of the Space Shuttle Columbia lost in early 2003, was unveiled at NASA Ames on October 26. It has achieved a sustained performance of 51.9 trillion operations per second, or teraflops, making it one of the fastest supercomputers in the world, outperforming Japan's Earth Simulator, previously the world's fastest supercomputer. (IBM's Blue Gene/L last week clocked 70.7 teraflops and now ranks first.)
Jim Taft, who leads the terascale applications group in NASA's Advanced Supercomputing Division, said access to Columbia was determined by a NASA review committee that established a list of prioritized activities.
"The UCSC work was at the top of the list, so we were authorized to give early access to a number of UCSC projects. These guys have been burning up the cycles ever since," Taft said.
The NASA exhibit at SC2004 will feature work by several UCSC researchers, including simulations of gamma-ray bursts and nuclear combustion in supernovae by Stan Woosley, professor of astronomy and astrophysics, and his collaborators.
"Using this machine is like being allowed to take several spins around the track at the wheel of the winning car at the Indy 500. We are ecstatic about the results we are getting," Woosley said.
In the supernova study, Woosley is working with postdoctoral researcher Mike Zingale and researchers at Lawrence Berkeley National Laboratory (LBNL) to study in minute detail how a nuclear combustion front, or "flame," moves through a star as it explodes in a supernova. John Bell, Marc Day, and Chuck Rendleman are combustion scientists at LBNL who provided computer code for simulating flames that Zingale applied to the astrophysics of supernovae.
"It is novel having combustion scientists and astrophysicists working together this way, and it is an indication of the difficulty of the problem," Woosley said. "To our knowledge, this is the largest, highest resolution study ever done of nuclear combustion in a thermonuclear supernova."
Type Ia supernovae are the brightest thermonuclear explosions in the universe, as bright as ten billion suns, and they have become important as "standard candles" used to measure the size and expansion of the universe. The explosion begins as a few hot spots near the center of a white dwarf star, generating far more energy than convection or diffusion can dissipate. Carbon and oxygen fuse to form heavier elements, chiefly iron, and the temperature rises to ten billion degrees Kelvin. Since the ash is lighter than the surrounding fuel, bubbles form that float away, burning violently as they go.
Over the next second, most of the star is consumed, releasing enough energy to explode as a supernova. The properties of the explosion, including its brightness, are determined by the rate of nuclear fusion during that critical second, and a calculation based on "first principles" has eluded astrophysicists for decades because the flame is subject to a variety of instabilities that are tricky to model. The range of length scales is also enormous, from the one-millimeter thickness of the flame to the 2000 kilometer radius of the star.
Woosley's group is simulating floating bubbles bounded by a carbon fusion front propagating in an essentially infinite reservoir of fuel. Only a small piece of the entire supernova is carried in the calculation. These simulations are quite complex and only a massively parallel computer with a lot of memory is capable of doing the calculation. On Columbia, the researchers are able to run this simulation long enough to see the complexities in the evolution develop. This simulation has already grown so large that they would have been unable to carry it out on any other machine, Woosley said.
"The complexity of NASA's space-derived data have become so great that tools of this sort need to be developed to make sense of it all," Woosley said. "We expect numerical simulation will increasingly dominate theoretical astrophysics in the coming years. For better or worse, we have moved beyond the ability to solve the most important problems using pencil and paper alone."
Brandon Allgood, a graduate student working with UCSC professor of physics Joel Primack, is at the NASA exhibit at SC2004 this week to describe the dark matter simulation he has been running on Columbia. A video derived from the results of the simulation will be shown at the NASA exhibit.
"This simulation is being used to understand the formation histories and current shapes of the dark matter halos around galaxies," Allgood said.
Dark matter--mysterious particles that make up at least 90 percent of the universe--has shaped the evolution of the universe through its gravitational pull on the ordinary, observable matter of galaxies and stars. Galaxies formed within large halos of dark matter, and the clumping of dark matter also guided the formation of galaxy clusters and other large-scale structures in the universe.
Primack has been using dark matter simulations to model the evolution of structure in the universe for many years. His group's latest simulation completed its run in less than a week using just a fraction of Columbia's processing power. Primack said he had been unable to get enough time to run the simulation on a Department of Energy supercomputer, where it would have taken over a month to run the same code.
"We're very excited to have early access to this supercomputer. We are now running an even bigger and more ambitious project than the one that will be on display at the conference," Primack said.
Other UCSC projects running on the Columbia machines are led by Gary Glatzmaier, professor of Earth sciences, and Piero Madau, professor of astronomy and astrophysics. Glatzmaier's group is studying complex processes in the deep interiors of stars, like the Sun, and large "gas giant" planets like Saturn. Madau and graduate student Michael Kuhlen are studying the early evolution of the universe with a simulation that follows the cosmological coevolution of dark matter and primordial gas.
Columbia is a highly integrated cluster of supercomputers driven by 10,240 Intel Itanium 2 processors. It comprises 20 identical nodes, each with 512 processors and one terabyte of shared memory. The benchmark performance of 51.9 teraflops was achieved using 20 nodes, and Columbia's peak performance is rated at 61 teraflops. The Columbia system was integrated in 120 days and the supercomputer is now fully operational with more than 650 users.
For more information about the Columbia supercomputer, see www.nas.nasa.gov/About/Projects/Columbia/columbia.html.
For more information about SC2004, see www.sc-conference.org/sc2004/.
Note to reporters: You may contact Woosley at (831) 459-2976 or firstname.lastname@example.org, and Primack at (831) 459-2580 or email@example.com.
Images are available on request from Tim Stephens in the UCSC Public Information Office at (831) 459-4352 or firstname.lastname@example.org.