An ARMAX/GRACH time series model for IP traffic trace

Large-range dependence (LRD) is essential phenomena both in LAN and WAN data traffic. Modeling such traffic traces is significant to understand the nature of the original traffic and to synthesize simulation traffic traces. Existing work such as multi-fractal wavelet model (MWM) was proposed to model IP traces and has reached much exactness. However, the nature of IP traffic exploits burst that is not fully captured by the proposed models. This paper proposes a new time series model ARMAX/GARCH to exploit even more burst network traffic. Through experiment analysis, our model shows that ARMAX/GARCH is able to identify all the important statistical properties of IP traffic traces. By applying partition function and the multi-fractal spectrum as metrics to evaluate both MWM and our model, our model reveals superior than MWM model does while the moment order is positive.