Point Process Models of 1/f Noise and Internet Traffic

TL;DR

This paper introduces a simple point process model that captures the long-range autocorrelations and 1/f noise characteristics observed in internet traffic, based on a power-law distributed file size assumption.

Contribution

The model demonstrates that long-range correlations in internet traffic can be explained without considering network-specific properties or inter-packet timing.

Findings

Reproduces long-range autocorrelations in web traffic

Captures 1/f noise power spectrum

Independent of network and inter-packet timing

Abstract

We present a simple model reproducing the long-range autocorrelations and the power spectrum of the web traffic. The model assumes the traffic as Poisson flow of files with size distributed according to the power-law. In this model the long-range autocorrelations are independent of the network properties as well as of inter-packet time distribution.