A new structure-preserving method of sampling for predicting self-similar traffic

Abstract

The paper presents a structure-preserving method of sampling self-similar traffic with an application to network monitoring and resource provisioning. Based on the observation of the self-similarity of Internet traffic, we propose a new sampling technique (so-called the maximum-based sampling). We show that the resulting data suits perfectly for predicting the bandwidth required by upcoming traffic so that the resource provisioning can be done efficiently and intelligently especially for the context of IP over WDM networks. Hence, we prove mathematically that the proposed technique preserves the self-similarity of the traffic. Besides, experimental results using real Internet traffic show that unlike other sampling techniques (systematic sampling and stratified random sampling), the maximum-based sampling capture faithfully the traffic self-similarity. In order to assess the effect of the sampling technique impact on the performance of the traffic prediction,we undertake a series of prediction experiments using sampled traffic with the proposed technique and the other sampling techniques. A neurofuzzy model (? _SNF), the AutoRegressive Integrated Moving Average model (ARIMA) and the Linear Minimum Mean Square Error (LMMSE) are considered in this study for bandwidth prediction. Our experiments results show that the maximum-based sampled traffic--used for the identification of the prediction model--is the most suitable for predicting the traffic for different time scales.