Personal tools
You are here: Home Publications Trace-Based Parallel Performance Overhead
Document Actions

F. Wolf, A. Malony, S. Shende, and A. Morris (2005)

Trace-Based Parallel Performance Overhead

In: High Performance Computing and Communcations: First International Conference, HPCC 2005, Sorrento, Italy, September 21-23, 2005. Proceedings, ed. by Laurence T. Yang, Omer F. Rana, Beniamino Di Martino, Jack Dongarra, chap. Lecture Notes in Computer Science, pp. 617, Springer-Verlag GmbH.

Tracing parallel programs to observe their performance introduces intrusion as the result of trace measurement overhead. If post-mortem trace analysis does not compensate for the overhead, the intrusion will lead to errors in the performance results. We show that measurement overhead can be accounted for during trace analysis and intrusion modeled and removed. Algorithms developed in our earlier work [5] are reimplemented in a more robust and modern tool, KOJAK [12], allowing them to be applied in large-scale parallel programs. The ability to reduce trace measurement error is demonstrated for a Monte-Carlo simulation based on a master/worker scheme. As an additional result, we visualize how local perturbation propagates across process boundaries and alters the behavioral characteristics of non-local processes.

by admin last modified 2007-12-10 21:06
« September 2010 »
Su Mo Tu We Th Fr Sa
1234
567891011
12131415161718
19202122232425
2627282930
 

Powered by Plone

LACSI Collaborators include:

Rice University LANL UH UNM UIUC UNC UTK