Optimizing Edge-Cloud Server Selection: A Multi-Objective Deep Reinforcement Learning Approach

Edge computing presents an efficient solution for alleviating the workload on cloud servers. However, effectively selecting edge and cloud servers poses a significant challenge. Numerous factors influence the performance of this server selection problem, including delays and link utilization. This paper addresses the issue of server selection in edge networks and cloud server clusters. Specifically, a Multi-Objective Deep Reinforcement Learning (MORL)-based resource scheduling scheme that integrates Proximal Policy Optimization (PPO) is proposed. Furthermore, we introduce a meticulously designed state encoding method and a sophisticated reward function to enhance the accuracy of utility computations. Experimental findings underscore the efficacy of the MORL scheme, demonstrating substantial improvement compared to existing server selection algorithms.