Efficient SDN-Based Traffic Monitoring in IoT Networks with Double Deep Q-Network

In an Internet of Things (IoT) environment, network traffic monitoring tasks are intractable to achieve due to various IoT traffic types. Recently, the development of Software-Defined Networking (SDN) enables outstanding flexibility and scalability abilities in network control and management, thereby providing a potential approach to mitigate challenges in monitoring the IoT traffic. In this paper, we propose an IoT traffic monitoring approach that implements deep reinforcement learning technique to maximize the fine-grained monitoring capability, i.e., level of traffic statistics details, for several IoT traffic groups. Specifically, we first study a flow-rule matching control system constrained by different expected levels of statistics details and by the flow-table limit of the SDN-based gateway device. We then formulate our control optimization problem by employing the Markov decision process (MDP). Afterwards, we develop Double Deep Q-Network (DDQN) algorithm to quickly obtain the optimal flow-rule matching control policy. Through the extensive experiments, the obtained results verify that the proposed approach yields outstanding improvements in terms of the ability to simultaneously provide different required degrees of statistics details while protecting the gateway devices from being overflowed in comparisons with those of the conventional Q-learning method and the typical SDN flow rule setting