| With the rapid development of information technologies, the number of Internet users dramatically increases, and network scale expands rapidly. The structure of networks also becomes complex and changeable, so network management and maintenance also become particularly difficult. It is difficult how to obtain the global network traffic information without affecting the normal operation of networks. And to perform effective resource optimization, network maintenance, fault diagnosis and so on according to the information has become the enormous challenges that Internet development faces. Because network traffic can use traffic matrix to describe network traffic state, we usually adopt the traffic matrix for reconstruction and estimation. It can completely describe network traffic matrix in the distribution of all the end-to-end network traffic demand. To get traffic matrix is not easy. However, because to attain the end-to-end network traffic by the direct measurement is not feasible, we build a reconstruction model able to accurately and effectively describe network traffic characteristics. Through the model to reconstruct the network traffic with the future tendency, it has become an effective way to get traffic matrix.In view of the problems of highly undetermined and pathological characteristics in the process of network traffic reconstruction, and the self-similarity of network traffic, polychotomy traffic characteristic, we put forward the network traffic reconstruction model ASMG based on multi-time series analysis theory. ASMG algorithm respectively reconstructs the high frequency fluctuated part and low frequency stable part of the traffic with different methods. We estimate the high frequency part by using the AR SaS model in time domain values reconstruction and the low frequency part by using MA-GM method. The reconstruction model can better describe the characteristics of network traffic compared with the traditional traffic models, and improve the precision of the reconstruction of the network traffic. Thus it can accurately reconstruct the real network traffic. Aiming at the multi-time series analysis model poor local characterization ability problem, and the length correlation of the network traffic itself, time-varying non-stationary and space-time correlation traffic characteristic, network traffic modeling and reconstruction of FRFT-WT method is proposed based on Fractional Fourier transform and wavelet transform. The reconstruction model adopts the wavelet decomposition to divide the original network traffic into high frequency and low frequency part. The high frequency part with short correlation and non-stationary uses fractional Fourier transform model to reconstruct. The low frequency part with long related characteristic adopts FAFIMA model to reconstruct. The final synthesis estimation is from the two parts, and final reconstruction results are obtained.With the multi-time series analysis model poor local characterization ability problem, FAFIMA model global grasp ability insufficient problem, and the uncertainty, non gaussian and some statistical properties of the network traffic, ICAWPT network traffic modeling and reconstruction algorithm is proposed based on independent component analysis theory and wavelet packet analysis theory. The reconstruction model, respectively, uses the two different methods to with reconstructand estimate the unknown network traffic. Finally, we carry on the geometric weighted integration, and realize accurate reconstruction of the network traffic. |
