Cross-architecture prediction based scheduling for energy efficient execution on single-ISA heterogeneous chip-multiprocessors

Abstract

Listen

Translate article

In recent years, single-ISA heterogeneous chip multiprocessors (CMP) consisting of big high-performance cores and small power-saving cores on the same die have been proposed for the exploration of high energy-efficiency. On such heterogeneous platforms, an appropriate runtime scheduling policy lies at the heart of program executions to benefit from the processor heterogeneity. To date, most prior works addressing this problem concentrate on the performance enhancement; however, they lack detailed justification of the runtime energy consumption and do not result in the most energy-efficient execution all the time. In this work, we pay attention to reducing the energy consumption for workloads running on heterogeneous CMPs and propose a scheduling algorithm based on dynamic execution behaviors to exploit better energy-efficiency. Our strategy is capable of significantly reducing the energy consumption while delivering comparable performance to a recently proposed heterogeneous scheduler (MLP-ratio), thus improving the energy-efficiency impressively.