| Measurement and analysis of traffic traces are important for better understanding of network behaviour. In this research, we collected traffic traces from a hybrid satellite-terrestrial network operated by ChinaSat, a commercial satellite Internet service provider.;We performed traffic analysis on packet, connection, protocol, and application layers. We investigated self-similar and long-range dependent characteristics of the collected traffic traces by employing various estimators. A traffic model on the TCP connection level was proposed. The relative fitness of various distributions functions was investigated. We concluded that Weibull and lognormal distributions are ideal for modelling the TCP inter arrival times and the number of downloaded bytes, respectively.;We then used billing records data to obtain one-week traffic prediction using autoregressive integrated moving average model. We implemented and evaluated the predictability of the redundant wavelet transform and combined it with autoregressive model to perform short-term traffic prediction. |
