Stable Matching Scheduler for Single-ISA Heterogeneous Multi-core Processors

The overall performance of single-ISA heterogeneous multi-core processors (HMPs) heavily relies on the efficiency of scheduling algorithm. However, traditional scheduling algorithms either treat all cores equally, or assume all cores complying with a strict order in the amount of microarchitecture resources, thus it is unsuitable for increasingly diverse HMPs, where different cores may have different advantages and preferences. To efficiently schedule single-ISA HMPs, we propose a novel stable matching scheduler based on the matching game theory. The proposed scheduler can always lead to a stable matching between applications and cores, where there does not exist an alternative application-core pair in which both the application and the core prefer each other rather than their current partners. Experimental results demonstrate that the stable matching scheduler beats previous schedules. For example, the stable matching scheduler improves, on average, the performance against the random scheduler by 19.86 % (4-core), 18.78 % (6-core), 22.78 % (8-core).