Research And Implementation On Traffic Classification Toward Application Features

With the development of network technology, the scale of traffic becomes larger and larger. Different types of applications presents a huge difference in the demand of network resources. Therefore, in order to guarantee the QoS demand for applications and achieve the refinement of management, it is necessary for traffic classification and identification. However, on the one hand, existing traffic classification methods are to identify the type of application based on protocol characteristics, they usually can not reflect the usage of the network resources and difficult to provide support for personalized management decisions. On the other hand, existing traffic classification method is limited by computing and storage capabilities of the platform in the face of large-scale traffic data. Thus, this paper includes the following aspects according to the influence of traffic features of the network application:(1) According to the analysis of existing application traffic QoS requirements and the usage of network resources, traffic features is introduced and the model toward the traffic features is established to characterize the dynamic impact of the type of application behavior on the network status, so as to guarantee the QoS demand for applications and achieve the refinement of management.(2) This paper proposes an algorithm for feature selection based on feature reduction in order to solve the problem of the excessive cost of training. Firstly, based on the categories features are selected based on the correlation on flow classification, then remove some features according to the high redundancy. At last, feature set is obtained which is help to reduce training time.(3) This paper proposes an algorithm for machine learning combined with Spark and SVM in face of the limited storage capacity and long training time. Firstly, several subsets of samples are divided using by under-sampling and oversampling method. Then, it uses Spark computing framework to construct the classifiers toward the subset of samples respectively. Finally, network traffic classifiers vote for the traffic classification.In summary, this paper studies the problem of traffic classification and proposes a model of traffic classification based on traffic features. The method contains feature selection and traffic classification in Spark. The experiment simulation and the implement of the prototype system are to verify the feasibility and effectiveness of the result.