Bringing Paxos Consensus in Multi-agent Systems

Reaching consensus has long been regarded as one of the most important problems in distributed systems. Being able to do so under failures is addressed by the Paxos family of algorithms which is able to guarantee safety, while probabilistically satisfying progress. The role-based approach of Paxos makes it an ideal candidate for implementation using autonomous agents which can dynamically assume their part(s). This paper aims to bring the basic Paxos fault tolerant algorithm into the multi-agent world by describing a system architecture and implementation under the Jade platform, and presents valuable experimental results for convergence in unfavorable cases.