Precise prediction of network traffic makes great contributions to grasping the network running trends and avoiding network failure. Aiming at the problem of low accuracy and slow convergence in the long-term network traffic prediction, a Wavelet Coefficient-aware Network Traffic Prediction (WCNTP) mechanism is proposed. By using Rescaled Range(R/S) sequence analysis, the statistical characteristics of network traffic on the large time scale are evaluated. Then the non-stationary network traffic is decomposed into a number of relatively stable network traffic sequences by discrete wavelet transform. Finally, the network traffic is predicted by using the Fractional Auto-Regressive Integration Moving Average(FARIMA) model. Results show that, the proposed mechanism has high accuracy and fast convergence speed in the process of long-term network traffic prediction, by which the network performance can be evaluated accurately, thereby improving the network service quality and ensuring the smooth operation of the network.