Adaptive threshold-based energy-efficient scheduling algorithm for parallel tasks on homogeneous DVS-enabled clusters

Increasing attention has been directly towards the energy efficient scheduling algorithms for parallel applications in high performance clusters. The existing duplication-based energy scheduling algorithms mainly leverage a threshold to balance system performance and energy consumption. However, the threshold is given randomly which cannot flexibly adapt to the cluster system and application performance requirements, thus making the ideal energy efficient scheduling results. In this paper, we propose a novel two-phase Adaptive Threshold-based Energy-efficient Scheduling algorithm (ATES). At first, we propose an adaptive threshold-based task duplication strategy, which can obtain an optimal threshold. It then leverages the optimal threshold to balance schedule lengths and energy savings by selectively replicating predecessor of a task. Therefore, the proposed task duplication strategy can get the suboptimal task groups. Then, it schedules the groups on the DVS-enabled processors to reduce processor energy whenever tasks have slack time due to task dependencies. The algorithm combines DVS(Dynamic Voltage Scaling) technique with adaptive threshold-based task duplication strategy. It justifies the threshold automatically to improve the energy efficiency of the scheduling algorithm. To illustrate the effectiveness of ATES, we simulate the real-world applications and compare ATES with the other four common task scheduling algorithms. Extensive experiment results show that our algorithm can much effectively balance schedule lengths and energy savings.