Research On Network Resource Allocation Based On Machine Learning In Software Defined Network

With the rapid development and popularization of the Internet,the scale of the network has continued to expand,and a large number of new types of network applications and services have emerged,such as Web search,Vo IP,video conferencing,IPTV,online gaming,and live webcasts.Different applications and services have different requirements for performance indicators such as delay,jitter,and packet loss rate in network transmission.Compared with the traditional Internet,the software-defined network(SDN)implements a more flexible and controllable data flow management method by decoupling control and data planes of network devices,which not only provides flexible management and deployment methods for various new types of networks,also helps to achieve differentiated and guaranteed service quality.Based on the new SDN network architecture,it is currently an important research topic to rationally allocate network resources and improve service quality and network performance.This thesis proposes a data-driven network resource allocation scheme in the SDN network.It identifies the QoS type of traffic entering the SDN network through QoS-aware traffic classification,and uses QoS adaptive routing algorithm based on reinforcement learning for routing allocation according to the traffic QoS type and real-time network states.The main research work of the thesis includes:(1)For the problem of poor real-time performance and poor generalization of single classifier in traffic classification methods in SDN networks,a QoS-aware traffic classification method based on SDN networks is proposed.The statistical characteristics of the traffic are extracted from the first 10 packets entering the network as the basis for classification.The improved integrated semi-supervised machine learning ITri-Training-3 algorithm is used to train the classifier,which implements real-time online recognition of network flow types and improves the recognition accuracy.In addition,combined with the DPI technology achieves a periodic incremental update of the online classifier,improves the practicality of the traffic classification method.(2)Aiming at the problem that the cost of QoS routing calculation in the existing SDN network is higher,and the real-time states of the link is not considered,a QoS adaptive routing algorithm DQPSR based on reinforcement learning is proposed.This algorithm implements optimized QoS routing calculation by introducing softmax action selection strategy,Q-Learning value function updating method and Markov decision process with QoS-aware reward function.The algorithm can fast adaptation to the time-varying network and traffic states,can well distributes the network traffic loads and has a good ability to scalable learning and so on.(3)Based on QoS-aware network traffic classification and QoS adaptive routing algorithm DQPSR,a QoS network resource allocation framework in SDN network is designed,and this framework is implemented on the open-source Floodlight SDN controller.The simulated SDN network environment is built with the Mininet network simulator to test the performance of the DQPSR algorithm.The experimental result shows that the DQPSR algorithm can achieve convergence in an effective time,andhas better performance compared with the shortest path algorithm and the load balancing algorithm.