Linear Control Theory for Modelling, Designing, and Performance Evaluation of ATM Congestion Control Algorithms

Abstract

This paper exploits classical control theory to design congestion control algorithms for ``best effort'' traffic in ATM networks. The control goal is the full utilisation of network links without incurring cell losses. A fluid model approximation of cell flows is assumed and linear differential equations are used to model the dynamics of network queues in response to ABR traffic and quality constrained (CBR+VBR) traffic. The available ABR bandwidth is modelled by means of an unknown and bounded disturbance function. A general end to end control algorithm that feeds back to the source the space that is free in the network buffers is included in the model and a particular one is proposed which is based on Smith's principle. The linearity and the simplicity of the proposed control law allow us to prove, via mathematical analysis, that the algorithm always guarantees no cell losses whereas full utilisation of network links is ensured if the capacity of per-VC buffers is at least equal to the VC bandwidth-delay product. Moreover, per-VC queuing easily allows switches to enforce fairness. Finally, it is shown how performance evaluation can be easily carried out using SIMULINK for MATLAB, which is a software tool widely used by control engineers to simulate dynamic systems. In this way, the effort to develop discrete event simulations is saved.