Optimization of Computer Networks

Chapter 6Congestion Control Problems

6.1 Introduction

Communication networks are intended to carry the traffic offered by the sources. When traffic volumes are too high, network links saturate, and there is a situation called congestion. To prevent it, a mechanism is needed to perform a fair and effective sharing of the link bandwidth by assigning the traffic volume each source is allowed to inject.

The network elements taking the bandwidth sharing decisions traditionally receive different names in different contexts. When traffic demands are virtual circuits or traffic connections at a fixed rate, the bandwidth sharing is governed by an admission control logic that decides for each connection request whether to carry it or block it. In turn, if the target is adjusting periodically (e.g., at a subsecond rate) the traffic injected by a demand, such that network resources are not saturated, the process is called congestion control. This chapter is mostly devoted to congestion control in packet switched networks.

Congestion control algorithms are code running in the traffic sources that computes an upper limit to the allowed rate to inject in the network. Some sources do not have much traffic to transmit and the congestion control upper limit is actually not restrictive for them. These are typically called mouse connections and correspond to occasional traffic between standard users. In contrast, other sources are always willing to transmit as much traffic as they are allowed. They ...