Research On Load Balancing Strategy Based On Yen Algorithm In SDN Network Architecture

With the rapid development of 5G,cloud computing,and the Internet of Things,traditional network architectures are no longer suitable for data centers.Software definition network(SDN)is different from the traditional network architecture.Because the control plane and forwarding plane separation are separated by the controller to implement the unified management of the network,and the advantages of openness and programming are gradually applying widely.There are two types of flows in the data center: one is an elephant flow with a small number and a high bandwidth;the other is a mouse flow with a large number and a small band occupation rate.The elephant flow in the network is concentrated in the local location of the network.It is easy to cause network congestion.Therefore,it is important to detect the elephant flow in the data center network and reasonably scheduling.In the framework of SDN,this paper studies elephant flow detection and load balancing in data center network.In this paper,an adaptive elephant-flow detection module is designed in the research of elephant-flow detection under SDN network architecture.In the first detection stage of the module,a statistical method is adopted to temporarily set the detection threshold and preliminarily detect the suspicious elephant flow in the network.In the second detection stage,a random forest(RF)and support vector machine(SVM)fusion RF-SVM detection model is proposed to screen out the real elephant flow.In order to improve the detection accuracy,firstly,the rat stream with obvious data characteristics is filtered out in the first detection stage to reduce the data processing capacity of the second detection stage model.Then,in the RF-SVM model of the second stage of detection,the advantage of RF that can process large-scale data in a highly parallel way is further screened to eliminate the data easily misclassified in SVM,and the remaining data is input into SVM to classify the real elephant flow.In terms of shortening the detection time,particle swarm optimization(PSO)was used to optimize the parameters of SVM,and the amount of data sent into the detection model was controlled according to the network state to reduce the detection time of the model.In this paper,the load balancing model of Yen-M elephant flow is designed to detect the elephant flow under SDN network architecture.The model uses Yen algorithm to calculate the K-short path.Meanwhile,the shortest path selected by Yen algorithm of machine learning module is the optimal path with the best network status.The path weight in Yen algorithm is an important parameter in its shortest path iteration,while the relaxation factor and fault tolerance parameters in SVM classification can control the precision of classification.The Yen-M model formed by combining the two can not only ensure the physical shortest link,but also ensure that the shortest path calculated by the model is optimal in real time and has the least overhead.In the process of detection experiment,five performance indexes such as detection accuracy and accuracy were used to evaluate various detection models.The experiment proved that the self-adaptive elephant flow detection module designed in this paper can realize high-precision and low-delay elephant flow detection.In the load balancing experiment,the average network throughput,average transmission delay and average link utilization rate are used as performance indicators to measure various load balancing models.The experiment proves that the Yen-M model designed in this paper can improve the load balancing efficiency of elephant flow in data centers and improve the network bandwidth utilization under the SDN architecture.