Novel Approach of Computational Resource Allocation in Fog Computing Based on Deep Reinforcement Learning Strategies
Degan Zhang, Jie Zhang, Xuemei Zhu, Ting Zhang, Xiumei Zheng, H. Jia
Abstract
The mobile Internet of Things (IoT) has gained popularity due to the quick advancement of mobile communication and intelligent terminal technologies.Focusing on some computationally demanding activities and latency-sensitive services (like health IoT) in smart healthcare that cloud computing (CC) cannot process and respond to rapidly. This research examines the fog computing (FC) and deep reinforcement learning (DRL) strategy-based edge computing resource allocation technique. This methodology generates computational tasks for mobile users at random throughout time. The mobile user has the option to load these tasks to the fog node at the edge or carry out local activities on additional mobile devices (MDs). A proximal policy optimization approach based on DRL is proposed for allocating computational resources to achieve low latency and low system energy consumption, thereby maximizing system revenue while concurrently enhancing the system’s overall quality of service. Its core idea is to combine the advantages of FC’s low-latency processing at the edge with the adaptive decision-making capability of DRL for dynamic resource states. This is achieved by considering the processing location of the computational tasks and the interaction between the device that generates the tasks and other MDs or FC nodes (FNs). According to experimental data, this approach has successfully decreased energy usage and network latency, compared with other algorithms. When the number of candidate nodes is 3, it achieves the lowest average latency and a notably reduced average energy consumption. Moreover, as the number of MDs increases, it maintains the optimal total system overhead and yields the highest average revenue.