In distributed simulations, many entities (mobile agents,
	distributed objects, processes) communicate and may evolve continuously.
	Volumes of communications, groups of communicating entities, as well as
	computational powers needed may vary a lot during the  execution. Due to
	these dynamic characteristics, the need of a migration for one or several
	entities is generally difficult to evaluate. This paper presents a method
	for computing such evaluation during the execution. A dynamic communication
	graph constitutes the model for the multiagent system. The proposed method
	evaluates the traditional trade-off between communication overhead and
	load-balancing by identifying clusters of highly communicating entities. A
	variant of classical ant algorithms was designed for that purpose. In our
	implementation pheromones are colored. The vertices of the communication
	graph are colored according to the effective communication between each
	other. A change of color for a vertex informs that a migration of the
	corresponding entity on the corresponding processor should be beneficial.
	Preliminary results demonstrate the interest of such an approach.