DELIGHT: a willingness-aware collaborative edge service offloading utilizing deep reinforcement learning

Under the impetus of mobile edge computing, mobile access networks and the Internet are deeply integrated providing low-latency, high-reliability computing power for service innovation. Due to the limitation of computing power resources, it is necessary for mobile smart terminals to offload essential service requests to edge servers to effectively enhance service quality. With the profound effect of selfishness, edge devices always make decisions to maintain their own performance regardless of the overall situation of the edge network. To this end, this paper proposes a willingness-aware collaborative offloading method in a mobile edge environment. It aims to efficiently offload services according to the offloading willingness of mobile smart terminals and the service willingness of edge servers. This approach seeks to resolve the selfishness in edge service offloading while reducing offloading costs and balancing the load on edge servers. At first, This method conducts the edge service offloading willingness model in a mobile edge environment. Subsequently, it models the willingness-aware service offloading problem in a mobile edge environment and transforms it into a multi-objective optimization problem with multiple constraints. Finally, a Markov decision process is constructed by utilizing DDPG, and a low-cost, high-quality, and load-balanced offloading scheme is ultimately achieved through iterative optimization. Experimental results on the Shanghai (Beijing) Telecom mobile communication base station dataset and the Shanghai (Beijing) taxi trajectory dataset indicate that the method proposed in this paper can effectively obtain willingness-aware service offloading schemes in mobile edge environments, and it outperforms typical baselines such as Random, GA, IA, DQN in overall performance.