This paper explores the application of state-space methods to congestion and delay control in communication networks. In the absence of buffer under- or overflow, the transmission delay for packets moving through a congested link turns out to be a non-linear output of a system with linear dynamics. As a consequence, the congestion/delay control problem can be addressed, in the presence of input saturation, by combining a disturbance feedforward plus state feedback control with an observer designed to cope with measurement delays. Both the feedback and observer designs are state-space versions of a Smith predictor.