Research And Implementation Of Network Performance Analysis System Based On Traffic Characteristics

With the rapid development of network technology,the types and number of applications running in modern networks are becoming more and more,and network performance has become the focus of attention.Through the analysis and prediction of traffic data,it can help network providers to clearly understand whether the provided network has reached the expected target of the network design,and accurately grasp the network traffic and performance in the future,which makes it more advantageous in competition with other network providers.The traditional traffic prediction models are not suitable for predicting time series with self-similarity,and the prediction accuracy of modern traffic prediction models is not high.Based on the traffic data set and performance indicator data set provided by ZTE,this thesis mainly does the following research work:(1)Aiming at the problem of whether the traffic has self-similarity,the R/S graph method is used to calculate the Hurst value at different time scales of the same node traffic,which proves that the traffic data has self-similarity.The traditional linear model can not accurately predict the traffic sequence,and the nonlinear time series model can be used for prediction.(2)Aiming at the problem of traffic prediction with self-similarity,this paper proposes an echo state network prediction method based on grid search.This method uses the ridge regression learning algorithm to solve the pathological problem caused by the traditional linear regression algorithm learning output connection weight matrix.The grid search method solves the parameter optimization problem of the reserve pool parameter and the regularization coefficient,and avoids the heuristic algorithm may fall into the local optimum when searching for the optimal parameters.The superiority of the model is proved by comparison with the basic echo state network,support vector regression and Elman models..(3)Aiming at the prediction problem of broadband utilization performance indicators,since the traffic prediction is essentially equivalent to predicting the byte throughput of one of the network performance indicators,the prediction model is extended to the prediction of the broadband utilization index,and the validity of the model is proved by experiments.In addition,based on the analysis of traffic characteristics,the effects of self-similarity,suddenness and periodicity on network performance are analyzed.(4)The network performance analysis system based on flow characteristics was developed,and the research results of this thesis were successfully applied to it.