K-Nearest Neighbor Search in Peer-to-Peer Systems

Data classification in large scale systems, such as peer-to-peer networks, can be very communication-expensive and impractical due to the huge amount of available data and lack of central control. Frequent data updates pose even more difficulties when applying existing classification techniques in peer-to-peer networks. We propose a distributed, scalable and robust classification algorithm based on k-nearest neighbor estimation. Our algorithm is asynchronous, considers data updates and imposes low communication overhead. The proposed method uses a content based overlay structure to organize data and moderate the number of query messages propagated in the network. Simulation results show that our algorithm performs efficiently in large scale networks.