Predicting the Upper Bound of Web Traffic Volume Using a Multiple Time Scale Approach

This paper presents a prediction algorithm for estimating the upper bound of future Web traffic volume. Unlike traditional traffic predictions that are performed at a single time scale using curve fitting, we employ a multiple time scale approach and utilize traffic statistical properties to do forecasting. We have applied our prediction algorithm to the 1998 World Cup data set. Experiments show that it is effective for short term traffic bound predictions, applicable to bursty traffic, and useful for Web server overload prevention.