Energy-Efficient Heterogeneous Computing of Parallel Applications via Power Capping

High-performance computing (HPC) systems require significant energy to run their operations. A typical HPC system contains various heterogeneous hardware components, including CPUs and GPUs. Power capping is a widely-used feature in processors to achieve an upper limit of power allocation. In this paper, we exploit power capping capability in modern processors to achieve energy-efficiency in heterogeneous computing system. We first develop an optimal power cap allocation model considering heterogeneous computing platform. Next, we develop a simulator based on a parallel discrete-event simulation engine to simulate our proposed power capping allocation model. Finally, we perform trace-based simulation experiments to demonstrate effectiveness of our model. Experiments demonstrate that our proposed model is capable of achieving various level of energy-to-solution reduction for different parallel applications considering heterogeneous computing platform.