Empirical Performance Modeling of HPC Workloads
Performance Analysis and Optimization
Scientific Software Development
TimeWednesday, June 27th2:07pm - 2:29pm
DescriptionMany parallel applications suffer from latent performance limitations that may prevent them from scaling to larger problem or machine sizes. Performance models allow such issues to be predicted before they become relevant. A performance model is a formula that expresses a performance metric of interest such as execution time or energy consumption as a function of one or more execution parameters such as the size of the input problem or the number of processors. However, deriving such models analytically from the code is so laborious that too many application developers shy away from the effort. To let a wider audience of developers profit from performance models, we created techniques to learn them automatically from a small set of performance measurements. In this talk, we present Extra-P, a performance-modeling tool that generates such empirical performance models for each function of even complex applications with hundreds of thousands of lines of code. We show the latest enhancements of Extra-P and illustrate its functionality in case studies.