Towards Adaptive Power-Aware Scheduling for Real-Time Tasks on DVS-Enabled Heterogeneous Clusters

Nowadays, power reducing has become a critical issue in clusters for minimizing electricity cost, improving system reliability and protecting environment. Consequently, developing power-aware scheduling strategies for applications on clusters, especially on heterogeneous clusters is highly desirable. We propose in this paper an adaptive power-aware scheduling strategy called APAS for a periodic real-time tasks on DVS-enabled heterogeneous clusters. APAS takes the scheduliability, power consumption, and system load into consideration. While scheduling, APAS is capable of adaptively adjusting voltage levels according to the system load. When the system is in heavy load, APAS increases voltage levels to improve scheduliability. In contrast, when the system is lightly loaded, APAS degrades voltage levels to reduce power consumption while guaranteeing high scheduliability. Compared with Greedy, SLVL and SHVL, APAS shows obvious excellent scheduling quality to others by simulation experiments.