FP-Outlier Mining Based Malicious Nodes Detection Model For P2P Networks

P2P network solves single point failure and performance bottleneck problems of classical Client-to-Server model, provides anonymous protection and guarantees of service quality of nodes in the network with node connectivity and file sharing in a peer-to-peer way. In recent years, P2P network has access a wide range of development in file sharing, multimedia transmission, distributed computing, collaborative work areas, etc. There are three kinds of P2P network:centralized, fully distributed and semi-distributed. Fully distributed P2P network has huge bandwidth consumption and poor scalability issues. Centralized one has problems of single point failure and load imbalance. Compared to the above structure, semi-distributed P2P network has been widely used in practice because of the advantage of both:high scalability, balanced load, efficient management, etc. Issues related to semi-distributed P2P network have also become a hot topic. P2P network provides the conditions for the malicious nodes to start its attack while provides a convenient and efficient service for users since it has the features of open, anonymous, self-organization. File pollution, bonnets as the representative, The P2P network malicious attack seriously affected the performance and development of the network. Reducing the impact of malicious attacks and improving the safety of P2P networks have become an urgent problem to be solved. Current most research uses trust mechanism to secure P2P network, select high quality services by evaluating the credits of nodes, so as to avoid the insecurity sources. The general method to obtain credits of nodes is mutual evaluation and recommendation between nodes, although this method improves the network security to some extent, the computing process relies on feedback and recommendation information, therefore has a bad evaluation performance in a large-scale network with sparse feedback, especially when malicious nodes offer false feedbacks, conspiracy attack, Sybil attack, etc. against the trust model.Essentially, the research purpose of improving network security is to predict which kind of service nodes will provide with history data, therefore avoid the malicious attack and select the quality service. In P2P network, there are lots of interaction data of nodes neglected by previous studies, such as interaction frequency, upload amount, which reflecting the behavior patterns of nodes in networks. These data are very important in the assessment of behavioral patterns of the interaction nodes in the past. When a malicious node lunches an attack, it will has distinct behavior patterns from other normal nodes no matter which kind of attack it starts, which makes the behavior patterns of malicious nodes exhibit the merits of a outlier in the network.This paper, based on outlier mining in data mining area and objective data interaction between nodes, proposes a malicious node detection model which is suitable for semi-distributed P2P network from the perspective of data analysis. Firstly, construct a behavior pattern model with the interaction data between nodes in networks; secondly, extract the local frequent behavior patterns in P2P subnets using frequent pattern mining method; Thirdly, this behavior pattern is more general than local frequent behavior pattern, and also can protect some P2P subnets from controlling by groups of malicious nodes; Finally, analyze the relationship between node behavior pattern and local frequent pattern, node behavior pattern and global frequent pattern respectively, obtain the local outlier factor and global outlier factor of node, evaluate the malicious level of node and P2P subnet. Simulation results show that this method can effectively identify malicious nodes and subnets in P2P networks in less time and space complexity.