NCSA Home
Contact Us | Intranet | Search

NCSA NEWS

News Home
Calendar
Images
Video on Demand
Subscribe to Our Newsletter
Frequently Asked Questions

PECM staff aims to analyze software performance

Performance analysis is half art and half science. It requires understanding the applications and algorithms of a discipline as well as the nuances and intricacies of hardware architectures, operating systems, and utilities. The results can be informative, groundbreaking, and just plain interesting, especially if presented in a clear, concise graphical format.

NCSA's Performance Engineering and Computational Analysis (PECM) staff, part of the Scientific Computing Division, work on analyzing performance on NCSA systems as well as on systems located at other sites. The purpose is to recognize the inherent advantages some systems have for a type of computation and to acknowledge the innovative uses that can develop from new approaches to old problems. Cross-discipline use of specific techniques is one way to break new ground in high-performance computing.

Graphs recently unveiled by the Performance Analysis team within the PECM staff summarize several key categories and trends among major research applications at NCSA. Because there are many ways to compare the macro performance of applications and their interactions with the underlying architectures, the team had to refine its choices. The results are online at http://www.ncsa.uiuc.edu/SCD/Perf/Tuning/.

"We can compare single processor performance of an application across several machines, or we can measure several applications on the same machine," notes project leader Mark Straka. "We can similarly measure parallel scalability, either on one machine or many. The possibilities sometimes seem endless."

Compilers and operating systems are constantly evolving and being upgraded, so the team tries to represent the current and best performance and highlight that on the Performance Analysis web pages.

Results from other teams are encouraged. Researchers are invited to contribute their results to Straka via email at straka@ncsa.uiuc.edu. For assistance in optimizing a code, see http://www.ncsa.uiuc.edu/SCD/Perf/Tuning/Form.html.

 

More News Briefs