AI-Based Virtual Network Function Embedding for Internet of Things Using Parallel Deep Reinforcement Learning

Although network function virtualization has provided a breakthrough in realizing the agility and flexibility for network resource management, virtual network functions (VNF) embedding is a challenging task in the Internet of Things (IoT) environments due to their specific requirements such as Quality of Service (QoS) and energy awareness. While there is a lot of research on VNF embedding on its own, there is very little on the load-balanced, energy-aware solutions. In this paper, we propose a load-balanced energy and delay aware deep reinforcement learning framework based on the advantage actor–critic learning strategy for efficient embedding of VNFs by considering the QoS and energy requirements of the IoT applications. Compared with the deep deterministic policy gradient method and the so-called DDRA (deep Q-learning-based dynamic resource allocation) method, our proposed approach shows outstanding performance in the successful acceptance rate of requests.