A Deep-reinforcement Learning Approach for SDN Routing Optimization
Chunlei Xu, Weijin Zhuang, Hong Zhang
Abstract
In order to realize a real-time and customizable routing optimization for software-defined network (SDN), the deep reinforcement learning method is integrated to the routing process of SDN. In this paper, we design a novel routing optimization mechanism based on deep reinforcement learning. This mechanism is capable of reducing the network maintenance and operational costs via improving network performance (e.g. delay, throughput), and it can also realize black-box optimization in continuous time. In addition, we conducted a series of experiments to evaluate the proposed routing optimization mechanism. The experimental results demonstrated that our proposed routing optimization mechanism has good effectiveness and convergence, which can not only achieve more stable performance than the traditional routing mechanism, but also provide better routing schemes.