Internet Traffic Modelling and Forecasting.

Internet traffic modelling and forecasting approaches have been studied and developed for more than decades. Most of the current proposed Internet traffic models focus on capturing traffic characteristics for ease of simulation and queueing analysis, but ignore models accuracy includes traffic source behaviours and possible policing or congestion avoidance. Since the current Internet traffic applications pay more attention to traffic engineering and QoS guarantees, which need accurate traffic source models to describe and forecast traffic behaviours and provide possible policing and congestion avoidance. So that, new traffic modelling applications such as accurate forecast traffic source behaviours need to be introduced into Internet traffic engineering tasks. However, classical traffic models, which can capture Internet traffic characteristics, do not have strong traffic performance forecast ability. Hence, this research focuses on the design and evaluation of innovative effective Internet traffic forecasting models. First of all, a new Internet traffic modelling and forecasting approach is proposed. This approach is based on wavelet multiresolution analysis with linear time series analysis. Wavelet multiresolution analysis is a wavelet analysis technique that simplifies the Internet traffic complexity and isolates the traffic’s long term trend with variability at multiple time scales. Linear time series analysis is a time series analysis technique with strong prediction ability. Evaluation results show that the proposed approach can achieve good prediction performance for real Internet traffic traces. Next, a new Internet traffic predictor with conditional variance characteristic is proposed. This model uses linear time series ARIMA process to model the traffic trend and adopts conditional variance structure GARCH process to model the ARIMA process’s innovations. In theory, the GARCH model can capture time series with high burst characteristics. Compared with the fractional based FARIMA model, the proposed predictor achieves better prediction performance. Finally, a new traffic predictability definition mechanism is proposed. This mechanism is developed based on real Internet traffic prediction applications. The predictability of the four most significant traffic predictors is studied. Analytical results show that these traffic predictors have different predictabilities for different traffic prediction conditions