We consider simulations constituted by numerous entities which
	evolve and interact dynamically. To be able to rep- resent a signi¿cant
	problem as an ecosystem, the modeliza- tion must be distributed. We are in
	front of a problem where the processes must obtain resources in a dynami-
	cally changing environment and be designed to collabo- rate despite of a
	variety of asynchronous and unpredictable changes. In such a context,
	applications are computational ecosystems analogous to biological
	ecosystems and human economies. This analogy suggests to tackle the problem
	as a complex system, in particular for the emergent aspects and the
	self-organization. Thus we detect organizations by an ant algorithm which
	allows to take into account the dy- namic side, load balancing of the
	processes and minimiza- tion of the impact of the communications between
	them. This has been tested successfully on a modelization of a very simple
	aquatic ecosystem where organizations appear, evolve and disappear.