Fast learning of optimal connections in a peer-to-peer network

Peer-to-Peer (P2P) protocol design is widely undertaken with the assumptions that peers and network connections are homogeneous resources. In practice this assumption is untrue. Furthermore, while there are some P2P networks that provide asymptotic cost optimal topology, or routing, very few existing protocols combine topology optimization and resource optimization, the resulting performance is resource oblivious. We propose a class of traffic based learning protocols, called FLOC protocols, that learn new connections between neighbors of neighbors. For an average routing table size of d and n peers, our protocol is seen to quickly converge from an inefficient network to an asymptotic cost optimal diameter of O(logd n), and simultaneously reduce network delay by approximately 50% in highly heterogeneous networks. We provide extensive simulation results to show FLOC´s behavior.