Energy-Efficient Phase-Aware Load Balancing on Asymmetric Multicore Processors

Comparing to the traditional symmetric design, single-ISA asymmetric multi-core processors were shown to deliver higher performance at lower energy costs. In such systems, load balancing is a challenging problem. Traditional algorithm is based on the load contribution prediction according to the task priority and the CPU utilization, ignoring the different characteristics of phases within a single task. Such approach might lead to suboptimized load balancing decisions, especially when concerning energy efficiency. In this paper, we propose a novel phase-aware load balancing method. Based on empirical data, proposed method could distinguish different execution phases. Comparing to HMP load balancing, the evaluation results show that the energy cost can be saved by 4% using the proposed approach while the CPU frequency is dynamically changed. If the CPU frequency is fixed, the energy cost can be saved by 10%.