At present, most of the popular P2P topology sharing systems are based on unstructured P2P topology. This topology uses flooding method to spread query information, which has good stability, but the efficiency is very low. According to the changing characteristics of the communication attribute state of the network behavior which occupies a large network bandwidth and communicates more frequently, an attribute migration state-oriented network communication behavior analysis method is proposed. The research on the identification of P2P traffic is of great significance for the management of P2P network. In view of the shortcomings of the current P2P traffic identification methods, such as large error and unstable identification results, in order to improve the identification effect of P2P traffic, a P2P traffic identification method based on neural network is proposed. This paper proposes a network file sharing system Kapa, which is based on a hybrid hierarchical P2P topology, which combines the advantages of unstructured and structured P2P topology and has strong practicability. The recognition effect and reliability of this method for P2P applications are verified.