Lucky Scheduling for Energy-Efficient Heterogeneous Multi-Core Systems

Heterogeneous multi-core processors with big/high-performance and small/low-power cores have been proposed as an alternative design to improve energy efficiency over traditional homogeneous multi-cores. We make the case for proportional-share scheduling of threads in heterogeneous processor cores aimed at improving combined energy efficiency and performance. Our thread scheduling algorithm, lucky, is based on lottery scheduling and has been implemented using Linux performance monitoring and thread-to-core affinity capabilities at user-level. Our preliminary results show that lucky scheduling can provide better performance and energy savings over state-of-the-art heterogeneous-aware scheduling techniques.