EDITORS: Energy-aware Dynamic Task Offloading using Deep Reinforcement Transfer Learning in SDN-enabled Edge Nodes

Abstract

In mobile edge computing systems, a task offloading approach should balance efficiency, adaptability, trust management, and reliability. This approach aims to maximise resource utilisation, improve user experience, and satisfy application-specific requirements while taking into account the dynamic and limited resource nature of edge environments. Additionally, while offloading tasks, these systems are vulnerable to several attacks and privacy breaches, necessitating edge node trust evaluation. However, not all of these necessary features are present in the offloading methods currently used. This research proposes ‘EDITORS’ (energy-aware dynamic task offloading method utilising Deep Reinforcement Transfer Learning (DRTL) in Software-Defined Network (SDN) enabled edge computing environments), a novel approach aimed at addressing the multifaceted issues associated with offloading in mobile edge computing systems. The primary goal of EDITORS is to design a task offloading system that incorporates trusted edge nodes while prioritising energy efficiency, timeliness, reliability, adaptability, and outperforming existing task offloading methods in terms of the quality of the task offloading plan. This method uses of DRTL agents at edge nodes, which communicate with the SDN controller to learn the most appropriate offloading choices based on network conditions and resource availability. Extensive simulations (six) are conducted which show that the EDITORS significantly increases energy efficiency while preserving low-latency task completion compared to five existing offloading methods (DDLO, DROO, DMRO, DRL without TL and SDN and DRL with SDN). EDITORS includes trust evaluation, trusted edge device prediction using LSTM, and adaptation of newly added devices through transfer learning, unlike other task offloading methods that just concentrate on task offloading.