Research On SDN Network Traffic Measurement Method

With the rising of network business requirements and increasing network scale and complexity,network operators need real-time control network performance parameters,including flow matrix,network of delay,packet loss rate of the network traffic,etc.The distributed control in the traditional network makes the operators have weak control ability of the network and can not achieve accurate network traffic measurement.Software definition Network(Software Defined Network SDN)as a new Network architecture arises at the historic moment,it is the most important idea is to spin out the Network control plane and transport plane,compared with the traditional Network SDN have stronger and more flexible central control ability,it can real-time distributed convection of fine-grained control flow chart,flow chart can obtain real-time flow statistics.SDN flexible network control ability for the precise measurement of the flow of each possible,but at the same time also has the problem of insufficient resources,in the SDN network limited resources of network traffic measurement is the key point of this study.Network traffic measurement under SDN is always an important research direction.In the data center,network resources(memory,bandwidth,and computing power)can be used in a coordinated way.How to allocate network resources for network collaborative measurement to solve the shortage of resources is the first problem we need to solve.We first proved that the problem was a np-hard problem,and then we modeled the problem into an ILP(Integer Linear Programming)model and designed the LPSAM algorithm.LPSAM(Lagrange Place Selectors And Monitors)algorithm ueses Lagrange multiplier method to solve the problem and divide the original problem into many subproblems.We use the depth first search of pruning optimization to solve subproblems.LPSAM algorithm gives the upper and lower bounds of the optimal solution of the original problem.We improved the LPSAM algorithm and obtained the ILPSAM(Improve Lagrange Place Selectors And Monitors)algorithm.The experimental simulation shows that the ILPSAM algorithm can provide approximate solutions quickly,and is suitable for fast deployment measurement tasks.The network measurement of SDN is focused on the network measurement when the anomaly occurs,and the network manager often needs to understand the current network traffic matrix.Under the background of SDN,accurate measurement of a flow need to consume a tri-state content addressable storage(TCAM)resources,and SDN switches TCAM resources are limited,under the condition of TCAM resources limited flow measurement matrix is our second job.We first designed a TCAM resource allocation algorithm which ensures that all traffic is measured within the minimum measuring period.In addition,in order to measure the flow matrix in a measurement period,we designed a flow detection algorithm based on reinforcement learning.The NetworkRL algorithm detects which flows are the stream and only measures the flow.Experiments show that the performance of NetworkRL is 20% better than the existing algorithm MUCB.Current switch buffer scheduling strategy is treated equally in the network each flow,however,a small number of large flows of the switch caching,make the most of the small flow cannot be scheduling,traffic flow detection can help switches for fine-grained network traffic management.Based on the existing network flow measurement,we use the convolution neural network and flow characteristics of 100 ms before convection flow and rapid detection of small flow,the accuracy of the experimental simulation won the90 %,as to optimize the buffer scheduling policies of switches.In order to make the network manager aware of the network traffic type of the current network,we use clustering and classification algorithm to carry out traffic identification.The simulation experiment obtained the accuracy of about 90%,which indicates that it is feasible to use the existing network traffic measurement data for traffic identification.