Adaptive routing for P2P networks using reinforcement learning

In this paper we focus on building keyword search service over unstructured Peer-to-Peer (P2P) networks. Current state-of-the-art keyword search approaches for unstructured P2P systems are either blind or informed. Blind search methods such as flooding in Gnutella generate a large of redundant cloned messages and waste network bandwidth. Informed approaches such as routing indices can allow nodes to forward queries to neighbors that are more likely to have answers but it could not suitable to dynamic network. In order to acquire more target resources in short time, we propose an intelligent method based on reinforcement learning (RL). The proposed algorithm can select optimal paths to send/forward the query messages according to the estimated value of arriving time at the target, so the results can acquire as much desired resources as possible in the limited time. Experiment on the simulation proves that our method could be adaptive to topology changes and link state. Compared with random-walk system, it can dramatically reduce network traffic and computation time and can improve performance up 30%.