Research On Performance Classification Method Of Relay Nodes In Tor Anonymous Communication System

Anonymous communication technology is a communication technology that hides communication relationships through multi-level proxies,encryption,and obfuscation.Among existing anonymous communication systems,Tor anonymous communication system has the advantages of mature technology and ease of use,making it one of the most widely used anonymous communication technologies with the widest coverage of users.Since Tor uses multiple relay nodes to forward data,most of which are provided by volunteers with varying performance,it is not comparable to traditional network routers.This makes some users may choose low-performance relay nodes to build anonymous circuits,which in turn affects the communication quality.To address this problem,this thesis extracts effective statistical features based on the historical statistics of Tor relay nodes and the interrelationships among Tor relay nodes,and classifies Tor relay nodes using methods such as deep learning and label propagation algorithms to identify high-performance relay nodes for users’ use.The main innovative work in this thesis is as follows:First,we propose a method to estimate the real-time performance of nodes based on the historical statistics of Tor relay nodes,and construct circuits using the identified highperformance relay nodes to verify the validity of the estimation algorithm.This thesis extracts the historical time series of multiple parameters of relay nodes from Collect Tor service as features,interpolates the time points of missing data to obtain the data set for training GRU neural network,and then classifies the currently collected online relay nodes to select the high-performance relay nodes.The experimental results show that the performance of the selected high-performance relay nodes is significantly better than that of the Tor default routing algorithm in constructing anonymous circuits.Second,we propose a semi-supervised method to classify Tor relay nodes based on the relationship features between Tor relay nodes,and use the labels of a small number of existing relay nodes to quickly classify all running relay nodes.In this thesis,we extract the family relationship,contact relationship,geographic distance,AS hop distance and selection probability among relay nodes from Onionoo,GeoLite2 and CAIDA datasets,analyze the distance and similarity features among relay nodes,and use an optimized label propagation algorithm to identify high performance relay nodes.The experimental results show that the anonymous circuits are constructed using the high-performance relay nodes identified by this method,which can significantly improve the Tor communication performance.