A Methodology for Developing Simple and Robust Power Models Using Performance Monitoring Events

In this paper, the authors describe a methodology for developing simple and robust power models using performance monitoring events for AMD Quad-core systems running OpenSolaris. The basic idea is correlating power consumption of a benchmark program with its performance (a measure of performance monitoring events). By using applicable model selection and model assessment techniques, they developed a simple and robust 2-predictor linear power model, which was shown to predict the power consumption of a testing set of benchmarks with better than 95% accuracy on average after being trained on a different set of benchmarks.