This paper deals with the bandwidth sharing and dynamic rate control in a connection-oriented multi-source network from the multivariable control theory perspective. Multi-source networks are characterised by heterogenous delays associated with their flows. General methodology in such cases involves design of a single input control (based on the upper bound of the delay) which is shared among the participating sources with the help of fairness algorithms. However, most of these fairness algorithms are iterative procedures and do not take into account the impact of variable delays on bandwidth allocation. By approaching this task from the multivariable control theory base, proposed method gives the flexibility to design optimal rate control in correspondence to the delays and removes the requirement of an explicit fairness algorithm. A novel discrete multivariable model of network is derived by segregating the sources according to their delays which leads to the property that the rank of input matrix and the system dimension are determined by the number of distinct delays and the upper bound of the delay respectively. With suitable choice of weighing matrices in the optimal control design, required fairness and throughput are ensured in the network. © 2014 IFAC.