Search

The rapid development of computer hardware and software provides a suitable platform for machine learning, in which deep learning has become a breakthrough in machine learning technology in various fields in many disciplines. Some recent research efforts have focused on routing control based on deep learning. Therefore, this paper studies the problem of intelligent routing, and aims to propose an intelligent control strategy based on deep learning with the help of Software-Defined Network (SDN) and other new network technologies. The characteristics of SDN network that can easily obtain the network topology have laid the foundation for selecting different routing paths according to the different QoS levels of the flow. Nowadays, the routing modules in commonly used SDN controllers use the shortest path algorithm which is simple to implement and works effectively. However, the best path calculated by controllers may suffer from huge traffic load and result in congestion, and the controllers cannot learn from the previous experiences to intelligently switch to other paths. This paper present intelligent routing control strategy based on Deep Q-Learning (DQN) in SDN, which uses the Openflow to collect information from the network, and aggregates them to the SDN controller, and then uses DQN to generate the specific routing for forwarders.

The current Internet is an open global interconnected system and lacks of the systematic securitydesign. Most security issues of current Internet are due to the drawbacks of the original design oftraditional Internet. Besides, some optional security mechanisms are independent without enough cooperativemechanism. Therefore, it facilitates the network attacks and brings various security threatsto network services. In this paper we design a security management mechanism and several relevantevaluation methods under Universal Identifier Network (UIN) architecture, aiming to provide a finergranularity, adaptive network security management system, which consists of the following features:(1) supporting multi-dimensional properties description by introducing the detailed user and serviceclassification; (2) supporting the label-based policy-driven management mechanism in perspective ofuser and service to provide the fine granularity access control; (3) providing the multi-dimensionalevaluation metrics. The proposed label-based security management and evaluation methods will providegreat benefits for the future network security.