Energy-Efficient Task Scheduling and Resource Allocation in Edge-Heterogeneous Computing Systems Using Multiobjective Optimization

With the explosive growth of Internet of Things (IoT) devices and the enormous data they generate, edge computing with heterogeneous architecture has been a research hotspot. Limited by a lack of power supply in the edge infrastructure, dynamic voltage and frequency scaling (DVFS)-based optimization becomes an effective way to balance energy cost and delay. However, achieving optimal performance efficiency in power-constrained edge computing environments remains a significant challenge, particularly in processing the tremendous requests from IoT endpoints. Therefore, we propose an energy-efficient task scheduling and resource allocation technique for edge heterogeneous computing. Specifically, we first utilize the directed acyclic graph (DAG) to describe IoT requests, and formulate task scheduling and resource allocation as a multi-objective mathematical programming model. Second, a Local Search enhanced Nondominated Sorting Genetic Algorithm-II (LS-NSGA) is designed as the solution method, where LS seeks to improve the problem solving capacity of NSGA-II. Additionally, we develop a strategy that automatically adjusts the DVFS setting for each processor in existing scheduling solutions, exploiting the potential energy reduction. Experimental results indicate that LS-NSGA outperforms existing state-of-the-art algorithms. In the best-case scenario, our method achieves 52% lower execution time and 73% lower energy cost than alternative approaches. Moreover, the ablation study shows that the proposed adaptive DVFS strategy can achieve additional energy savings of approximately 4% to 7% for various task-scale problems.