1 Nonlinear Instabilities in TCP - RED

This work presents a novel modeling paradigm of the dynamical negotiation between clients running TCP (Transmission Control Protocol) and routers with the RED (Random Early Detection) active queue management scheme. The basic aim of the modeling is to understand the inherent nonlinearity in the interaction and how it manifests itself in the parametric sensitivities observed in practice. The model proposed here is used to study network dynamics over large parameter variations. Both smooth bifurcations, such as period doubling, and nonsmooth bifurcations, such as border collision bifurcations, are shown to occur as system parameters are varied. The bifurcations, which involve the emergence of oscillatory and/or chaotic behavior, shed light on the parametric sensitivity observed in practice. Based on the observation of period doubling bifurcation as a basic instability initiation mechanism, we also propose a lightweight control algorithm to contain these instabilities.