Network traffic shaping based on prediction of polynomial trend self-similar time series

In the present paper shaping algorithms development is considered. Most attention is paid to shaping algorithms based on network traffic prediction. Estimates of prediction-based shapers efficiency for different forecasting techniques are obtained. It is shown that a shaping algorithm should take into account both the prehistory and future values of the traffic in order to achieve the maximum of its operation efficiency. The paper presents an adaptive linear predictor of the fractal network traffic and compares it to the simple autoregressive predictor. According to our simulation results, the autoregressive shaper grants significantly smoother output while the adaptive predictor grants significantly lower packet loss ratio.