Intelligent user-collaborative edge device APC-based MEC 5G IoT for computational offloading and resource allocation

Abstract

Mobile edge computing (MEC) supports delay-sensitive and excellent processing capacity services on the 5G Internet of Things (IoT) network. This research proposes an intelligent resource allocation policy to minimize average service latency and average energy consumption for an IoT system while maximizing the available processing capacity (APC). We express the APC as a function of communication and computation resources at the user and task edge devices. We demonstrated that reducing execution delay and energy usage could improve overall system service performance. We evaluate the savings in average latency and average energy consumption when we reduce execution delay by allocating resources to maximize APC. The 5G IoT network uses natural actor-critic deep reinforcement learning to tackle complicated resource allocation decisions, and the simulation shows that reducing execution time improves overall system performance. Our results improved execution time and energy usage compared to random search, greedy search, and deep Q-learning. In addition, our single centralized agent DRL outperforms Multi-agent DRL for the number of rewards and completed task achievable under different episodes.