Energy Efficient Resource Orchestration in Fog-Enabled Connected Vehicle Networks using Deep Reinforcement Learning

Abstract

Increase in technological advancements has led to a revolution in vehicular networks. Fog computing is integrated into the existing vehicular networks to form Fog-Enabled Connected Vehicle Networks (FCVN) to increase the latency and throughput. Orchestration schemes are proposed to solve load balancing issues with normal device association schemes. In particular, graph-based orchestration schemes have shown a remarkable increase in service capability. However, solving the orchestration problem with multiple objectives using iterative approaches is difficult and takes time to converge. To solve these problems, we propose an improvement to the existing graph-based orchestration schemes using Reinforcement Learning (RL). A robust reward function is also proposed to tune the importance given to different objectives. The maximum resource reduction possible between every pair of fog node (FN) is computed using RL and is considered as the weight of the link. Then, maximum weight matching is performed on the resultant graph to find optimal pairs of FNs for service migrations. To demonstrate the algorithm, two objectives, energy consumption and service capability is considered. The results reveal our algorithm simultaneously improves both the objectives when compared to the existing graph-based orchestration schemes.