An Ant Based System for Dynamic Multiple Criteria Balancing

Serge Fenet, Laboratoire  d'Ingénierie des Systèmes d'Information, Université Claude Bernard - Lyon 1, Villeurbanne, France
Email: s-fene95@bat710.univ-lyon1.fr

Salima Hassas, Laboratoire  d'Ingénierie des Systèmes d'Information, Université Claude Bernard - Lyon 1, Villeurbanne, France
Email: hassas@bat710.univ-lyon1.fr

Emergent behaviour in multi-agents systems is rooted in interaction. We study such a system applied to load balancing in a telecommunication network. We present in this work an ant-based multi-agents system that allows task migration over a computer network. The ant paradigm allows us to code information and constraints as pheromone fields. Thus, and because the computers load is coded with such a system, load balancing becomes a criterion among others. Our work present several advantages :

- a natural expression of the problem in terms of local interactions that evolve to a global equilibrium;
- the support of a non limitative number of criteria in the choice of the migratory path;
- the ability to set dynamicaly the relative importance of these criteria to respond to time-evolving requirements;
- a very quick adaptation to the disturbances of the network;
- the great robustness of a rective system and a very good response to local or global overloads.

At the crossroad of the Artificial Intelligence and the Network, our approach can lead to a new point of view on local networks and their fauna : the computationnal ecology.