Survey of traffic prediction methods for dynamic routing in overlay networks

We introduce and compare methods for traffic prediction focused on dynamic routing in overlay networks and their practical usage. Overlay networks often need special routing, even when underlay networks do routing on their own. There are more unknown variables then in traditional routing approach. Dynamic routing is based on several metrics and prediction could be judged by many statistical values. All introduced methods are compared by computing costs, implementation complexity and statistical values. Main statistical quantity is MAE (mean absolute error), which presents accuracy of prediction. We apply prediction on routing model to increase quality of service. So, we can focus on final route parameters not only on prediction accuracy. Naturally, main route parameter is transfer speed, but in real life scenarios we can find examples, where it is not only one metric. Let us assume real time audio/video stream, where we are interested in lowest range of scatter to reduce flapping of stream quality. In the paper, we will discuss impact of used metrics and algorithms on final quality of service.