Follow this link to skip to the main content
NASA - National Aeronautics and Space Administration

+ NASA Home
+ Ames Home

+ Sitemap
+ Staff Directory
+ NAS Internal


+Home


HIGH END SYSTEMS
+ Pleiades
+ Columbia
+ Schirra

SYSTEMS
APPLICATIONS
Application Optimization
VISUALIZATION
NETWORKS
SOFTWARE
GREEN

"Johnny provided us with diligent support in the setup and running of our simulation...we much appreciate the pleasant, informative, and very helpful communications with him."

— Hy Trac, NAS user


Two astrophysics researchers running simulations of the early universe on the Columbia supercomputer gained a 25% speedup in code performance, thanks to work by the NAS Division’s Application Performance and Productivity group.


Massive Astrophysics Simulation Benefits from NAS Code Optimization

In May 2008, researchers Renyue Cen, of Princeton University's Department of Astrophysical Sciences, and colleague Hy Trac at the Institute for Theory and Computation at the Harvard-Smithsonian Center for Astrophysics, benefited greatly from code execution and performance optimization support by Johnny Chang in NAS' Application Performance and Productivity (APP) group.

Cen and Trac are using their self-developed numerical code, based on N-body, hydrodynamic, and radiative transfer algorithms to model the process of reionization—where the first galaxies, dating approximately 13 billion years ago, heated and ionized the gas in the early universe through the radiation they produced.

map of temperature field of first galaxies

Above: A map of the temperature field halfway through this important epoch in the evolution of the universe. Reionization converts a cold and neutral gas into a warm and highly ionized medium.

Originally, the researchers' job was scheduled to run continuously for five days on Columbia. The users had several "start up" problems and the job ran much slower than expected. During the course of the run, the job was stopped multiple times, and was restarted with changes in the code checkpoint algorithm to resolve performance problems.

"Johnny provided us with diligent support in the setup and running of our simulations," Trac said, adding, "We much appreciate the pleasant, informative, and very helpful communications with him."

Chang’s extensive code support included set-up and running of Cen and Trac’s state-of-the-art simulation, which required large computer processor and memory resources—about 2 terabytes—on the Columbia supercomputer. He performed transfer and backup of important files and data, and diagnosed code performance problems. Chang also tracked down issues, ran experiments, and determined the cause of various problems. His suggestions for improvement sped up the code by 10%. Another suggested optimization—running the code on every other core—sped the code by another 15%.

Chang estimated that the collective changes would save two days of runtime (over an 8-day run) in the future. Cen and Trac commended Chang’s dedication, making himself available late at night and on weekends during a ten-day period to solve the problem, enabling the simulation to complete.

Contact: Johnny Chang, Johnny.Chang@nasa.gov, NASA Advanced Supercomputing Division, Computer Sciences Corp.




USA.gov -- government made easy
+ Feedback
+ Site Help
+ NASA Privacy Statement, Disclaimer, and Accessibility Certification
Click to visit the NAS Homepage
Editor: Jill Dunbar
Webmaster: John Hardman
NASA Official: Rupak Biswas
+ Contact NAS

Last Updated: March 27, 2009