Multi objective optimization of HPC kernels for performance, power, and energy

Abstract

Code optimization in the high-performance computing realm has traditionally focused on reducing execution time. The problem, in mathematical terms, has been expressed as a single objective optimization problem. The expected concerns of next-generation systems, however, demand a more detailed analysis of the interplay among execution time and other metrics. Metrics such as power, performance, energy, and resiliency may all be targeted together and traded against one another. We present a multi objective formulation of the code optimization problem. Our proposed framework helps one explore potential tradeoffs among multiple objectives and provides a significantly richer analysis than can be achieved by treating additional metrics as hard constraints. We empirically examine a variety of metrics, architectures, and code optimization decisions and provide evidence that such tradeoffs exist in practice.