Berkeley Researchers Win NSF PetaApps Award for CMB Applications
 |
 |
October 30, 2009
Contacts: Linda Vu, lvu@lbl.gov, 510-495-2402
 |
| This graphic shows variations in the cosmic microwave backgound as seen by by three different missions COBE, WMAP and Planck. |
By analyzing minute variations in the Cosmic Microwave Background (CMB), the remnant radiation from the Big Bang, cosmologists are expanding our knowledge of the universe. However, the CMB data that are being gathered by increasingly powerful experiments are also pushing the limits of today's most powerful computers. And it will only get more challenging–—experts predict that the volume of CMB data will increase 1000-fold over the next 15 years and push the limits for scientific computing.
To ensure that future computing systems meet the demands of the CMB community, researchers at the University of California at Berkeley will develop a new benchmark tool for testing whole-system performance on emerging extreme-scale supercomputing systems. The project is funded with a $755,604 award from the National Science Foundation's (NSF) PetaApps program, awarded to UC Berkeley under the American Recovery and Reinvestment Act.
"Every generation of supercomputer comes with a new challenge. A machine with more processors may calculate faster, but if the communication between processors is slow, then it is not going to work for many CMB problems," says Julian Borrill, a research physicist in UC Berkeley's Space Sciences Laboratory. "There are many disciplines with similar computing needs that will benefit from a benchmark that tests whole-system performance of a machine, giving you a more accurate picture of how the system handles a true scientific workload."
The project will occur in two phases. First the researchers will create a stripped-down test code that includes all the computational complexity, but none of the CMB scientific code base. This allows researchers to measure whole-system performance, identify bottlenecks, and implement solutions across a wide range of petascale computing systems. The lessons learned from this first phase will be incorporated into a CMB code suite that will automatically assess the strengths and weaknesses of the machine that it is running on, and self-tune to ensure that the scientific analysis is utilizing all aspects of the system efficiently.
"The most powerful supercomputers are scarce shared resources, and researchers typically compute wherever they can get time on a machine. This means that you may end up running your code on many different systems, each with different strengths and weaknesses," says Borrill. "This new code suite will automatically adjust for that so that you are utilizing the strengths of each machine and are getting the best science results possible. Extreme-scale systems will be an enormously valuable resource for scientific research, and it is crucial that we use them as efficiently and effectively as possible."
Borrill and Horst Simon, Adjunct Professor at UC Berkeley, are co- and principal investigators, respectively, of this NSF-supported project. Simon is also the Associate Laboratory Director for Computing Sciences at the Lawrence Berkeley National Laboratory (Berkeley Lab), while Borill is also a staff scientist in the Berkeley Lab's Computational Research Division's Computational Cosmology Center.
For more information about computing sciences at Berkeley Lab, please visit: www.lbl.gov/cs
