In this paper, we propose a general fluid flow network model based on non-linear diŽerential equations. In the case of a FIFO queue, we show that our model converges to the solution given by the queueing theory in term of queue length and queueing delay. We introduce a simple hop-byhop feedback control strategy and show that this method has a number of interesting properties particularly well suited to the control of single rate multicast flows. This problem is indeed known to suffer from scalability limitations which does not occur with a hop-by-hop approach. Stability and performance of hop-by-hop control methods have already been studied in the literature and results have shown that they perform well. The originality of this paper is to propose a model whose simplicity gives deep insight into the understanding of the dynamics of the system. In the unicast case, and for the special case of a chain of routers, this approach allows to demonstrate the boundedness of the buffer queues and the convergence of the system to a single globally asymptotically stable equilibrium. Furthermore, the proposed control method is implementation friendly.

Numerical simulations are used to illustrate the different properties of the control method. We then compare these results with experimental measurements. Our experimental platform is realized with User Mode Linux virtual machines. This method allows for low cost and realistic experimental validation. Furthermore the wide variety of network protocols and tools available for Linux are readily available in User Mode Linux.