Modeling and Prediction of the Internet End-to-end Delay using Recurrent Neural Networks

Abstract

This paper focuses on modeling and predicting the Internet end-to-end (e2e) delay multi-step ahead using Recurrent Neural Networks (RNNs). In this work, Round- Trip Time (RTT) is used as the basic metric to forecast the Internet e2e delay. A method for delay prediction model is developed using RNNs, able to model nonlinear systems. By observing the delay between two Internet nodes, RTT data has been collected as a time series during several days. Then this discrete-time series data has been organized into two parts, the first one is used as a training/learning set of the RNN, whereas the rest of data is used for the testing/evaluation of the RNN performance. To achieve this purpose, a learning phase has been performed to provide a mathematical characterization of RTT during one or several reference days. The test phase consists of iteratively forecasting RTT acquired during the test day. Simulation results illustrate that the suggested model is adaptive and it tracks RTT dynamics rapidly and accurately, even for long time ahead prediction.