On the correlation of TCP traffic in backbone networks

Abstract

In this paper, we study the second order statistics of traffic in an Internet backbone. We model the traffic at the flow level by a Poisson shot noise process. This model is quite parasimonious, and is driven only by variables that can be easily obtained from measurements, namely flow sizes, durations and arrival rate. We consider the auto-correlation of TCP traffic where the loss process of each TCP connection is assumed to be Poisson. Using a stochastic differential equation, we are able to provide an upper bound on the auto-covariance function of the aggregated TCP traffic whose tightness is shown by simulations with the network simulator-ns.

A Long-Range Dependent Model for Network Traffic with Flow-Scale Correlations

“…An application of this model is proposed in Ref. [38] where the shape of the shot is representative of the AIMD (Additive Increase Multiplicative Decrease) mechanism with Poisson losses. In Ref.…”