The present invention is a delay based model and in fact uses queuing delay as a congestion measure, providing advantages over prior art loss based systems. One advantage is that queuing delay can be more accurately estimated than loss probability. This is because packet losses in networks with large bandwidth-delay product are rare events under TCP Reno and its variants (probability on the order 10−7 or smaller), and because loss samples provide coarser information than queuing delay samples. Indeed, measurements of delay are noisy, just as those of loss probability. Thus, another advantage of the present invention is that each measurement of queuing delay provides multi-bit information while each measurement of packet loss (whether a packet is lost) provides only one bit of information for the filtering of noise. This makes it easier for an equation-based implementation to stabilize a network into a steady state with a target fairness and high utilization. In addition, the dynamics of queuing delay provides scaling with respect to network capacity. This helps maintain stability as a network scales up in capacity.