Research And Implementation Of SDN Flow Table Optimization Based On Machine Learning

With the continuous emergence of 5G,cloud computing and other emerging technologies,more and more complex functions have been added to the traditional IP architecture network.To some extent,this has led to the increasing overstaffing of switching equipment,the shrinking storage space,the increasing frequency of network congestion,and the difficulty in meeting the growing demand for high performance.Software defined network is considered to solve the IP network structure rigidity,improve resource utilization and one of the effective ways to promote the future network innovation,SDN network control plane and data plane,decoupling,control plane is responsible for the centralized control logic,data plane is responsible for simple data forwarding,and other functions,and by using TCAM(Ternary Content Addressable Memory)for storage.However,TCAM has disadvantages such as high energy consumption,high cost and limited capacity,so how to optimize the SDN flow table to maximize the utilization rate of flow table in the limited storage space is still an urgent problem to be solved at present.This paper summarizes and analyzes the current research status of existing elephant flow detection algorithms and SDN flow table optimization schemes,and then proposes an improved elephant flow detection algorithm based on stream mining and a timeout mechanism of flow table to distinguish elephant flow and mouse flow.In order to solve the problem of low accuracy of model detection caused by the imbalance of the proportion of elephant flow and mouse flow in the sampling process,the equilibrium sampling processing measures are added on the basis of the original stream mining algorithm.Then a dynamically weighted cost-sensitive detection method is proposed to improve the accuracy of elephant flow detection by avoiding misdetection and error-detection of elephant flow.Finally,after the detection of elephant flow,a timeout mechanism is proposed to distinguish elephant flow from mouse flow.First,most mouse flows are left in the switch through two-stage detection of the switch and the controller to reduce the communication overhead between the controller and the switches.The second is to accurately predict the idle timeout time of each type of flow by using the improved elephant flow detection algorithm.Thirdly,the timeout time of flow table entries is set dynamically and in real time by combining flow table utilization and predicted the range of flow duration time,and flow table entries are deleted proportionally before flow table overflow to improve flow table utilization.In order to verify the effectiveness of the proposed algorithm,simulation experiments were carried out in Mininet simulation environment respectively.Experimental results show that the proposed algorithm has better performance in terms of detection accuracy,communication overhead and flow table utilization compared with existing algorithms.