Phase-based scheduling and thread migration for heterogeneous multicore processors

Heterogeneous multicore processors (HMPs) can provide better performance and reduced energy consumption than homogeneous ones [3]. Differences between cores provide different processing capabilities for different applications; a dynamic scheduler can exploit these differences to maximize performance and minimize energy consumption [5, 6] by adapting to fine changes in programs behavior. This work proposes new fine-grained online HMP schedulers using program phase identification. However, exploiting fine-grained scheduling results in frequent thread migrations that can harm performance. OS context switching is time consuming (30-60μs [4]). To reduce context switching overhead, a context switching circuit that both accelerates thread switches among cores in HMPs and reduces switching cost within each core (multitasking) is introduced.