A Novel Non-Supervised Deep-Learning-Based Network Traffic Control Method for Software Defined Wireless Networks

Abstract

SDN has been regarded as the next-generation network paradigm as it decouples complex network management from packet forwarding, which significantly simplifies the operation of switches in the data plane. The good programmability of SDN infrastructure also improves network feasibility. To alleviate the burden of the explosive growth in network traffic, in this article we propose a non-supervised deep learning based routing strategy running in the SDN controller. In our proposal, we utilize the CNNs as our deep learning architecture, and the controller runs the CNNs to choose the best path combination for packet forwarding in switches. More importantly, in our proposal, the controller collects the network traffic trace and periodically trains the CNNs to adapt them to the changing traffic patterns. Simulation results demonstrate that our proposal is able to retain learning from previous experiences and outperform conventional routing protocols.