Agent-Based Approach for Network Allocation Problems

Numerous applications can be formulated as a resource allocation problem, which require a distributed solving method especially when dynamic environments are considered. According to such methods, a solution does not only consist in an optimal allocation, but needs a sequence of transactions leading from a given initial allocation to an optimal one. Up to now, studies have been based on ideal contexts. Indeed, agents are omniscient and/or have complete communication abilities. Most of the time, these assumptions are not plausible. We propose in this paper a multi-agent system in which agents elaborate themselves optimal allocations by means of local negotiations. This self-organized system is based on a more realistic context, where provided solutions can be viewed as emergent phenomena. Agents have a limited perception of their environment and restricted communication abilities. We show that the individual rationality, does not allow the achievement of socially optimal allocations, and we propose a more suitable criterion: the sociability. Our method provides a sequence of transactions leading to an optimal allocation, according to any communication networks.