Decentralized Consensus for P2P Network with Trust Relationships

The decentralized nature of P2P network increases robustness because it removes the single point of failure, however, reaching consensus over an open P2P network while maintaining decentralization is still an open problem. We proposed a consensus algorithm for P2P network based on trust relationships and social consensus with emergent phenomena of complex system. Starting from synchronous system model, we model the algorithm with mean field equations and analyze the performance of convergence as well as fault tolerance. We then adapt the algorithm to asynchronous system model by incorporating a message filter and a Chandra & Toueg style failure detector. Simulations show that on the SNAP dataset of the Wikipedia who-votes-on-whom network, the algorithm can always converge within 40 rounds in synchronous system model and 70 seconds in synchronous system model under reasonable latency assumption if without failure. Simulations also show that the algorithm can tolerant collusion of 15% random nodes or 2% top influential nodes committing attack of the strongest type.