Research On Analysis And Classification Of Network Traffic Based On Net Flow

With the further development of computer network technology and extensive application of Internet,people make more high-level demands of network performance and network security and the network planning and management become important increasingly.According to the specific data carried in the network,analyzing the network traffic in real time and understanding the trendency and features of traffic can provide decision data for network planning.Due to the fact that the methods orienting traffic data at low levels before is unable to satisfy the high-capacity and high real-time need of the network management at this stage,this paper relizes aggregation,classification,and prediction of traffic data based on NetFlow technology.And the development of NetFlow technology is becoming increasingly mature which has become an industry standard of monitoring,statistics,security management,accounting and so on of network traffic to some content.This paper puts forward a framework of traffic aggregation system based on NetFlow first.It collects data by nProbe,with pmacct storing the data in MySQL,the script aggregating the total traffic,national traffic and international traffic according to a certain granularity of time all day long as well as inserting them into a MySQL cluster and its frontend demonstrating the basic situation of the inflow and outflow of the traffic of the day,the week,the month and this year after aggregation in real time with webpage charts.The system has run in the real environment of the research institute,which can act as an effective tool for small-medium sized networks to implement the traffic statistics and elimination of network failures.Besides,the paper classifies the traffic in the module of data storage in the above system as national traffic and international traffic by C4.5 decision tree.It regards the top three segment of IP after data preprocessing as attributes,considering the attributes as branch nodes,attribute values as branches,classification labels as leaf nodes,where the test set walks along the model generated by the training set and reaches the leaf nodes in order to produce the classification result.The experiment result shows that the C4.5 algorithm taking the sample set after data compression and data augmentation as input possesses good performance on classification while the tree after pre-pruning by the roles guarantees the certain classification accuracy,which reduces the complexity of model effectively.Finally,the paper puts forward a kind of nonlinear regression model based on LM algorithm contraposing features of the numerical fluctuation of traffic by comparing with a linear regression model and a BP neural network,of which data is from the analysis module in the aggregation system.The nonlinear regression model based on LM requires the given initial value of parameters to be estimated in the nonlinear function,damping,mean square error and so on and iterates the weights continually by LM algorithm to solve the optimal solution,causing the result that the error value is less than the given value.The experiment result shows that the model is of accuracy and stability.