next up previous
Next: Bibliography Up: Performance Evaluation of Adaptive Previous: Case Study: Uintah


Conclusions

When studying the performance of scientific applications, especially on large-scale parallel systems, there is a strong preference among developers to view performance information with respect to their ``mental'' model of the application, formed from the structural, logical, and numerical models used in the program. If the developer can relate performance data measured during execution to what they know about the application, more effective program optimization may be achieved. In this paper, we present portable performance evaluation techniques in the context of the TAU performance system and its application to the Uintah computational framework. We illustrate how phase based profiling may be effectively used to bridge the semantic gap in comprehending the performance of parallel scientific applications using techniques that map program performance to higher level abstractions.



Sameer Shende 2005-05-30