Multi-Objective Evolutionary Algorithm Based on Decomposition for Energy-aware Scheduling in Heterogeneous Computing Systems

Sisi Yuan (Oklahoma State University, USA)

Gaoshan Deng ((Xiamen University, China)

Quanxi Feng (Oklahoma State University, USA)

Pan Zheng (Swinburne University of Technology, Sarawak Campus, Malaysia)

Tao Song (Swinburne University of Technology, Sarawak Campus, Malaysia)

Abstract: Heterogeneous computing systems (HCSs) use many heterogeneous processors or cores to perform particular tasks. To address the requirement of green IT, several power management techniques have been developed to reduce the energy consumption of these systems. Dynamic voltage scaling, which dynamically changes the supply voltage of processors during the execution of an application, is widely used. However, reducing supply voltage decreases computation speed. Therefore, system makespan and energy consumption need to be considered at the same time. We propose a multi-objective scheduling algorithm based on decomposition for scheduling of the system workflow. Through experiments, we examine the performances of several algorithms, including the proposed one, in different benchmarks and real-world applications. Results show that our algorithm demonstrates better performance than other state-of-art evolutionary algorithms under various conditions involving the use of different crossover and mutation operators.

Keywords: MOEA/D, energy efficiency, evolutionary algorithms, heterogeneous computing systems

Categories: D.4.1, H.1.2, H.4.0