The Research Of P2P-Oriented Network Traffic Trediction Technology

With development of computer technology in current society, the scale of the computer network also continuously been extended. The technology of P2P(peer-to-peer) has achieved rapid development and gradually occupied the majority part of Internet business, consuming a lot of network bandwidth. The vast flow of P2P network not only occupy the bandwidth of the other applications on the network, which hampers the conduct of normal network operation and the popularity of certain critical applications, but also can cause network congestion easily in the case of limited network bandwidth, greatly reducing the performance of the entire computer network, and serious degradation of network quality of service. Therefore, it is necessary to make research of P2P network traffic, in order to make full use of the limited network bandwidth and improve the quality of services for the entire network and optimize network traffic control and achieve the network reliable data transmission as well as the rational allocation of network resources.In this paper, through researching a variety of domestic and international traffic prediction model, analyzing respective advantages and disadvantages of some traditional traffic forecasting model and the new traffic forecasting technology and suitable, mainly make focus on analysis of the ESN (Echo State Networks) model structure and learning mechanism. Because it has the complexity of the network structure that is far superior to other neural network, the ESN have good way to nonlinear approach and could achieve good results in calculating nonlinear time series. However the prediction effect of ESN is not ideal, under the setting of the small sample size and the noisy, and ESN model can only solve the limited extent of the frequency range of multi-scale, have low adaptability of multi-scale time series prediction problem,is difficult to meet the high precision requirements of P2P traffic forecast.In this paper, I take advantages of the multi-scale analysis of wavelet transform, and combine with the advantages of ESN and propose an improved P2P traffic prediction model based on wavelet transform and ESN, by the reason of complex flow characteristics of P2P network, such as sudden, fractal self-similarity, periodic and chaotic etc. Firstly, through the wavelet decomposition and reconstruction to the original sequence of P2P traffic, you can get the high-frequency components and low frequency components of a series of different scales, the equivalent of the original P2P traffic smoothing, decomposition of the complex P2P network traffic relative to a single related component.Then we could embed dimension and weight of these different frequency components constitute a time series of network traffic through the phase space reconstruction theory to determine the optimal delay time and minimum. Finally, according to the reconstructed time series properties we can match the appropriate parameter ESN model training and prediction. Match the flow characteristics of the different components of the ESN with the model of the different parameters are predicted, and finally integration of the predicted results of the multiplex output.Collected different P2P software such as PPStream, QQLive, Thunder traffic forecasting results show that the prediction accuracy of the traffic flow prediction model proposed in this paper (mean absolute percentage error) can reach more than98%, significantly better than traditional ESN model and LS-SVM model.