Energy-aware Task Scheduling on DVS-enabled Heterogeneous Clusters by Iterated Local Search
Yujian Zhang, Yun Wang, Xin Yuan · 2018
Energy consumption in heterogeneous clusters has attracted lots of attention since it results in high operating cost and environmental pollution. Task scheduling is considered as an effective software approach to reduce energy dissipation on DVS-enabled clusters. However, single-objective optimization focusing on energy reduction lacks the consideration on makespan while recent bi-objective schedulers can hardly guarantee the performance on both makespan and energy-saving. In this paper, we present a bi-objective approach that can ensure the performance by iterated local search, namely ILS-DVS. The algorithm is started from an initial solution produced by a time-effective scheduler and explores more solution space by perturbation to find more feasible candidates. Then, a hill climbing method is employed to optimize these candidate solutions. After a certain number of iterations, a Pareto-efficient schedule can be produced by ILS-DVS. Experimental results demonstrate that the proposed algorithm can significantly improve both makespan and energy-saving, and has superior performance to other competitors.