An engineering approach to prediction of network traffic based on time-series model

Campus network's Internet accessing traffic is complicated, non-linear and periodical. Our goal is to give out a engineering approach to prediction of network traffic based Time-series analysis model(EPTS) for campus exit-link. In our EPTS with rate-limiting, we configure rate limit based interface, then use Time-series decomposed model, give out the linear trend component, periodical component, and random component decomposed analysis model. We analyze two years' traffic data of ECNU campus network exit-link and try to forecast the same link's traffic tendency of the following half year. We get the satisfied prediction results compared with either the link's real monitor data or Time-series analysis model without rate-limiting. We believe our approach is a feasible method for forecasting network traffic tendency.