Moving Average Forecasting for Network Traffic with Self-Similar
自相似网络通信量的滑动平均预测

Network traffic with heavy-tallness shows long-range burstiness which is totally different from conventional traffic in telephone network. Characterization and forecast for the long-range dependence traffic is very important for network performance analysis and network design. Two moving average predictors based on alpha-stable self-similar traffic model are presented. First predictor is a linear unbiased estimator based on covariation-orthogonality. Another predictor is asymptotically moving average forecast with symmetrical stable innovation and it can minimize the dispersion. Forecasting experiments for actual traffic trace from Bellcore Laboratory and Lawrence Berkeley Laboratory show that two predictors are accurate and reliable.