 |
Center for Nonlinear Studies |
Optimizing the performance of large-scale scientific simulations requires an understanding of the emergent behavior between the application software and the system on which it runs. Performance instrumentation allows collection of data about particular executions, in the form of aggregated profiles or detailed traces. However, interpreting the data collected is challenging. Visualization is often a part of the performance analysis process as there are many possible sources of performance degradation and understanding is necessary to make changes. The scale of the data and complicated nature of application behavior to represent tax most visual solutions. I will present performance visualizations that combine parallel semantics with performance data to alleviate this burden.
Host: John Patchett