Network environments become more and more complex in terms of new services, quality of service to handle and user demand for increasing throughput. All these facts have to be performed in real time (or almost), and consequently, network management and control are difficult to realise in this continuously changing environment. The network is now unpredictable and there is a real lack in the modelisation and simulation tools to handle this dynamicity. In this context, agents and multi-agent systems seem to be good candidates to provide a way to model and to manage the network dynamicity. This is due to the fact that agents can perform tasks in an autonomous, distributed, adaptable and cooperative manner. Our goal here is to model an active network by using a multi-agent approach and more precisely a behaviour based agent approach. Instead of representing the actions of the network elements by complex algorithms, we rather describe them by their behaviours which are usually simple and familiar words, inspired in part from human behaviour. On this basis, we have made simulations of an active network environment, and implemented different combinations of four individual behaviours by using a multi-agent platform called Swarm. Results of these simulations are significant and show that we do not loose important packets (with priority) any more when we introduce a behaviour in the nodes.