Predicting Internet End-to-End Delay: A Statistical Study

Predicting the end-to-end packet delay and understanding the Internet dynamics are of great importance for many realtime and nonreal-time applications over the Internet, especially for the quality of service (QoS) provisioning and congestion control. After some preliminary statistical analysis, this paper presents a novel approach to predict the Internet end-to-end delay using multiple-model (MM) methods. The basic idea of the MM method is to assume the system dynamics can be described by a set of models rather than a single one; by running a bank of filters (each corresponds to a certain model in the set) in parallel at the same time, the MM output is given by a combination of the estimates from these filters. Based on collected end-to-end delay data, we propose an off-line model set design procedure using vector quantization (VQ) and short-term time series analysis so that MM methods can be applied to predict on-line measurement data. Numerical results show that the proposed MM predictor outperforms two widely used adaptive filters in terms of prediction accuracy and robustness.