Online Internet traffic prediction models based on MMSE

Abstract

Traffic prediction model is critically important for network performance evaluation and services quality. Traditional traffic prediction models cannot reflect the characteristics of self-similar traffic. Current long-range prediction models, however, are too complex to be used as online traffic predictors. This paper presents two new traffic predictors which are MMSEP and NMSEP. They are based on minimum mean square error. Time series and control theory are used to build the mathematic models. By modifying the way of calculating the predicted error, MMESP and NMSEP can reflect the burst of self-similar traffic in multiple timescales. When compared with FARIMA model which is one of the best fractional predictor, numerical results of experiments show that MMSEP and NMSEP can achieve accuracy with less than 5% of errors while keeping simplify in computation and low memory used. © Springer-Verlag Berlin Heidelberg 2005.