Dynamic Placement Using Ants for Object Based Simulations Cyrille Bertelle Antoine Dutot Frédéric Guinand Damien Olivier Laboratoire d’Informatique du Havre Université du Havre 25 rue Philippe Lebon 76600 Le Havre email: Antoine.Dutot@univ-lehavre.fr Abstract. A distributed application may be considered as a set of interacting entities continuously evolving. Such application can be modeled as a graph with one-to-one mappings between vertices and entities and between edges and communications. Per- formances depend directly on a good load balancing of the entities between available computing devices and on the minimization of the impact of the communications between them. However, both objectives are contradictory and good performances are achieved if and only if a good tradeoff is found. Our method for finding such a tradeoff is new and based on colored ant colonies. Each computing resource is associ- ated to one ant colony characterized by a color, allowing an implicit consideration of the load balancing constraint. Then, using colored pheromones, ants are just seeking for communicating structures. The method operates on graphs which structural and numerical parameters may change dynamically during the execution. Keywords: Ant algorithms, dynamic graph, clustering, auto-organization, distributed applications . 1 Introduction In distributed application, often a very large number of entities are used to represent a complex system. The dynamics of such systems discourages a static distribution made upstream, before application execution. As the system evolves communications between entities change. Communications and entities may appear or disappear, cre- ating organizations. As a consequence, an entity location that was correct at the be- ginning, can severely impact performance two hundred time steps after. Therefore we need an anytime distribution method that advices the application on better locations for each entity preserving load-balancing between computing resources, but ensur- ing that entities that communicate heavily are close together (ideally on the same processing resource). In this paper, a method based on the Ant System[6] is described that advises on a pos- sible better location of some entities according to the tradeoff between load balancing and minimization of communications overhead. The Paper is organized as follows. Section 2 provides some background about ant algorithms and some of their applications. Section 3 details the graph representing the distributed application. Operating on this graph, our colored ant system is described in section 4. Finally, our implementation is discussed in section 5 and illustrated by some experiments, before we conclude with further expected improvements and perspectives for this system. 2 Ant Algorithms Ant algorithms are a class of meta-heuristics based on a population of agents exhibit- ing a cooperative behaviour[10]. Ants continuously forage their territories to find food[8] visiting paths, creating bridges, constructing nests, etc.