Supercomputers now aid scientific discovery on an unprecedented scale. Terascale computers analyze massive data quantities and create intensely detailed simulations, and in the next few years, petascale systems will boost those capabilities orders-of-magnitude further. Yet even the most powerful computers are hampered by their enormous size, increasing architectural complexity and proliferating, vastly more involved applications.
The Performance Engineering Research Institute for Enabling Technology (PERI), a $15-million, five-year coalition of 10 major research entities, is pursuing a tightly integrated approach to optimizing both systems and applications across the computing process. Led by ISI Division Director Robert F. Lucas and funded by the US Department of Energy’s Office of Scientific Computing, PERI is working to:
- Develop and refine performance modeling and prediction to reduce data collection costs, while increasing model fidelity and speed;
- Advance automatic performance optimization, or systems’ ability to analyze and tune themselves, to enable scientists to focus on research rather than code; and
- Optimize performance engineering of high-profile applications, including "tiger teams" that focus on particular codes.
Participants include many of the nation’s highest-profile research institutions:
Argonne National Laboratory, Lawrence Berkeley National Laboratory, Lawrence Livermore National Laboratory, Oak Ridge National Laboratory, Rice University, University of California at San Diego, University of Maryland, University of North Carolina, USC/ISI and the University of Tennessee. ISI’s participation focuses on automatic performance tuning.
For more about PERI’s goals and achievements, click here.