Implementation and analysis of the BitTorrent protocol with a multi-agent model

A peer to peer (P2P) computer network relies on direct connections between participants instead of conventional centralized resources. These networks are useful for many purposes, but the killer application nowadays is file sharing. BitTorrent is a very popular protocol for this matter. In this paper we model a BitTorrent network as a multi-agent system. In it, each BitTorrent client is an agent that interacts with other agents and reacts autonomously, following same decision algorithms as a real client. Their goals are downloading contents in the shortest time possible and sharing them with other agents. The model is programmed in JADE, a software framework in Java language for agent development. The model has been validated by initial tests, and it will be used to study the behavior of BitTorrent networks in diverse situations that are hard to emulate in a real network or with other models, like modifications of the protocol or behaviors of the users. Implementation and analysis of the BitTorrent protocol with a multi-agent model E. Costa-Montenegro, J.C. Burguillo-Rial, F. Gil-Castiñeira, F.J. González-Castaño Telematic Engineering Department. University of Vigo. 36310 Vigo, Spain Phone: +34 986813474 Fax: +34 986812116 Abstract A peer to peer (P2P) computer network relies on direct connections between participants instead of conventional centralized resources. These networks are useful for many purposes, but the killer application nowadays is file sharing. BitTorrent is a very popular protocol for this matter. In this paper we model a BitTorrent network as a multi-agent system. In it, each BitTorrent client is an agent that interacts with other agents and reacts autonomously, following same decision algorithms as a real client. Their goals are downloading contents in the shortest time possible and sharing them with other agents. The model is programmed in JADE, a software framework in Java language for agent development. The model has been validated by initial tests, and it will be used to study the behavior of BitTorrent networks in diverse situations that are hard to emulate in a real network or with other models, like modifications of the protocol or behaviors of the users.