Research On The Internet Traffic Identification

Traffic identification technology is a scientific research area which is very close to the practical industry technology and the application market. It requires that the researchers will not only have a comprehensive and in-depth understanding of the academic theory, but also take into consideration for the overall development of future Internet. Only in this way, these achievements can be helpful for the practical industry.This dissertation mainly focuses on the following three parts. First, it proposes a Unified Objective Ground Truth Generation(UOGTG) framework. Second, it investigates the real-time traffic identification technology based on flow-based statistic features. Third, it researches the feature generation technology of network traffic which can be applied to deep packet inspection. The major contributions and innovations of this dissertation are as follows.(1) This dissertation proposes a distributed traffic measurement based evaluation framework, which is called Unified Objective Ground Truth Generation Framework. It introduces distributed agents which collects different features from their local traffic and reports them to a converged server. It can be used for the evaluation of both the data mining based traffic identification algorithm and the automatic network traffic signature generation. UOGTG solves the lack of a reliable unified objective ground truth method in research area. The results indicate that this framework plays a fundamental role in further research of traffic identification research area.(2) This dissertation researches a real-time traffic identification model which can be used for flow-based data stream mining algorithm. Based on the intensive research on the features of network traffic, it proposed the first real-time traffic identification model for flow-based data stream mining algorithm. Moreover, it investigates the concept drift feature of network traffic, and proposes an ensemble-base algorithm to solve the concept drift problem.(3) This dissertation researches the traffic signature generation technology. It firstly proposes the framework for traffic signature generation algorithm, including several key technologies of signatures extracting, signatures management and signatures purification. And it also introduces two kinds of signature generation algorithms inspired by bioinformatics, which will also inspire the network traffic identification industry.