DISTRIBUTION OF AGENT BASED SIMULATION WITH COLORED ANT ALGORITHM

Cyrille Bertelle (1) , Antoine Dutot (1) , Frédéric Guinand (1) , Damien Olivier (1)

1 RI2C - LITIS - Equipe Réseaux d'interactions et Intelligence Collective

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 needs 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. 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.

Mots clés: Distributed Computing; Ant Algorithm; Simulation; Migration; Colored Pheromones

Domaines: Intelligence artificielle [cs.AI]

Cyrille Bertelle, Antoine Dutot, Frédéric Guinand, Damien Olivier. DISTRIBUTION OF AGENT BASED SIMULATION WITH COLORED ANT ALGORITHM. ESS 2002 - 14th European Simulation Symposium and Exhibition, Oct 2002, Dresden, Germany. ⟨hal-03317645⟩