In this talk, we begin with the progression of parallel computer system power management over the last two decades. In particular, we focus on the evolution of quantitative design approaches and the emergence of effective parallel and distributed system runtime power management. We observe that the growing complexity of today's machines limits the effectiveness of traditional quantitative approaches. The Compute-Overlap-Stall (COS) model of parallel computation is proposed to accurately capture the effects of emergent orchestrated power management of processor, memory, and thread throttling. The implication of our findings is that as power management techniques pervade, new machine-learning performance evaluation and prediction approaches will be essential to future computer system designs.

Another application for machine learning has been to create max power stressmarks. Manually developing and tuning so called stressmarks is extremely tedious and time-consuming while requiring an intimate understanding of the processor. In our past research, we created a framework that uses machine learning for the automated generation of stressmarks. In this talk, the methodology of the creation of automatic stressmarks will be explained. Experiments on multiple platforms validating the proposed approach will be described.

Yet another application for machine learning is in cross-platform performance and power prediction. If one model is slow to run real-world benchmarks/workloads, is it possible to predict/estimate the performance/power by using runs on another platform? Are there correlations that can be exploited using machine learning to make cross-platform performance and power predictions? A methodology to perform cross-platform performance/power predictions will be presented in this talk.