Research On Classification And Identification Of IPv6Network Traffic

Network traffic classification and identification is an important branch in the fieldof internet research. In-depth analysis of network traffic is of great significance fornetwork expansion and optimization, network security and analysis of upper layer users’behavior.In recent years, the continued evolution and development of the Internet havebrought new challenges to the research of network traffic classification. On the onehand, due to the complexity of applications, the traditional network traffic classificationbased on port and load has exposed its limitation. There is an urgent need for a moreeffective and more accurate network traffic classification method. On the other hand,the original IPv4protocol, due to its deficiencies, will be replaced by thenext-generation IPv6protocol. This renovation will also pose new challenges tonetwork traffic classification and identification. For the above problems, thisdissertation includes the following four parts.①According to the actual situation of IPv4and IPv6network traffic in ChongqingUniversity, the complete process of the new machine learning classification based onthe statistical characteristics of network traffic is introduced in this dissertation. Andtwo major difficulties faced in the IPv6network traffic classification are raised.②Since there is serious imbalance of application distribution in IPv6networktraffic, a classification method of multi-classifier ensemble learning is proposed in thisdissertation. Based on the network traffic classification with single machine learningalgorithm used by previous researchers, this method, combining with the features ofvarious classifiers, implements the network traffic classification by means of majorityvoting and instance selection. Experimental results show that this method improves theclassification accuracy and the generalization performance.③An improved DBSCAN clustering analysis method is proposed for unknownIPv6network traffic. Through deep analysis of the DBSCAN clustering algorithm basedon density, the time efficiency of this algorithm is improved. Finally the clusteringanalysis of campus IPv6network traffic is completed with this improved DBSCANalgorithm. Experimental results show that improved DBSCAN algorithm which candiscovery unknown traffic in the network gets better clustering results.④Most existing tools for traffic classification and identification only achieve one or several process in the classification. There is no complete machine learningclassification and identification tools for professional statistical characteristics ofnetwork traffic. In order to solve this problem, a prototype system for network trafficclassification and identification based on B/S mode is designed and implemented in thispaper. This system provides a good experimental platform for future work of networktraffic classification and identification.