DRL Based Service Migration and Resource Allocation in Vehicular Edge Networks

Future vehicular networks are expected to support various stringent delay-sensitive services of the vehicles using their widely distributed edge servers. However, the randomness in the mobility and service heterogeneity of the vehicles pose challenges in achieving the desired quality of service. Therefore, dynamic resource (i.e., bandwidth and computing resources) allocation and service migration across the edge servers become apparent. Unlike existing works, in this paper, each edge server independently performs proactive service migration and also allocates resources for the service migrations which could occur from other edge servers. Further, the bandwidth and computing resources are assigned to the vehicles in such a manner that it satisfies their heterogeneous service requirements. We formulate a model-free Markov decision process (MDP) to facilitate an edge server’s decisions. The objective of the MDP is to maximize the total expected utility of the services offloaded by the vehicles. The MDP is approximated and solved using the deep deterministic policy gradient algorithm (DDPG). We provide detailed analysis with comparative study to highlight the performance of the proposed model.