Multiscale Modeling and Queuing Analysis of Long-Range-Dependent Network Traffic

Vinay Ribeiro (vinay@rice.edu)
Electrical and Computer Engineering, Rice University

Rudolf Riedi (riedi@rice.edu)
Electrical and Computer Engineering, Rice University

Matthew Crouse (mcrouse@ece.rice.edu)
Electrical and Computer Engineering, Rice University

Richard Baraniuk (richb@rice.edu)
Electrical and Computer Engineering, Rice University

We develop a simple multiscale model for the analysis and synthesis of nonGaussian, long-range-dependent (LRD) network traffic loads. The wavelet transform effectively decorrelates LRD signals and hence is well-suited to model such data. However, traditional wavelet-based models are Gaussian in nature which one may at the most hope to match second order statistics of inherently nonGaussian traffic loads. Using a multiplicative superstructure atop the Haar wavelet tree, we retain the decorrelating properties of wavelets while simultaneously capturing the positivity and ``spikiness'' of nonGaussian traffic. This leads to a swift O(N) algorithm for fitting and synthesizing N-point data sets. The resulting model belongs to the class of multifractal cascades, a set of processes with rich scaling properties which are better suited than LRD to capture burstiness. We elucidate our model's ability to capture the covariance structure of real data and then fit it to real traffic traces. We derive approximate analytical queuing formulas for our model, also applicable to other multiscale models, by exploiting its multiscale construction scheme. Queuing experiments demonstrate the accuracy of the model for matching real data and the precision of our theoretical queuing results, thus revealing the potential use of the model for numerous networking applications. Our results indicate that a Gaussian assumption can lead to over-optimistic predictions of tail queue probability even when taking LRD into account.