Predicting End-to-End Delay of the Internet Using Time Series Analysis ∗

This report surveys time series methods that have been used and can be applied in predicting end-to-end delay of the Internet. ARIMA scheme and state-space approach are discussed and compared. Although state-space approach has the advantages in structure and computation, ARIMA modeling is still useful in identifying systems due to the complexity and uncertainty of the Internet. A practical example of using ARMA models to do prediction is given in the report. By converting the time series models (ARIMA or ARMAX) to state-space representation, linear prediction and control theory can be applied more directly. Furthermore, by regarding the network as a hybrid system, multiple-model approach, whose algorithms are mainly based on state-space representation and Kalman filtering, can be used to study the dynamics of the Internet.