Agent-Controlled Distributed Resource Sharing to Improve P2P File Exchanges in User Networks

摘要

In this paper, we evaluate the feasibility of distributed control of shared resources in user-managed networks. This paradigm has become possible with the advent of broadband wireless networking technologies such as IEEE 802.11. One of the most popular applications in these networks is peer-to-peer (P2P) file exchange. Node cooperation can optimize the usage of shared "external" accesses to the Internet (set of links between the user network and the Internet). In a previous paper, we evaluated different agent-oriented distributed control schema, based on the concept of credit limits, on ideal mesh networks subject to uniform traffic. Each node in the mesh network chooses to behave as a cooperator or a defector. Cooperators may assist in file exchange, whereas defectors try to get advantage of network resources without providing help in return. Before this paper was completed, we observed that popular P2P protocols like eMule, Kazaa and BitTorrent were evolving towards the same credit-oriented strategies we previously proposed. Now, we realistically model both user network traffic and topology, and evaluate a new advanced agent-based distributed control scheme. The simulation results in this paper confirm on realistic networks the main conclusion in our previous research: autonomous node agents become cooperators in their permanent state when they take decisions from local information, checking that file exchange services offered to neighbor nodes do not exceed appropriate credit limits.