Application Of Partition Weighted Incremental SVM In The Detection Of P2P Flow Traffic

In recent years, P2P technology as a new network applications, is leading the direction of development of the Internet. With the development of the P2P technology, P2P traffic has become one of the most significant portions of the network traffic with consuming large network bandwidth. Main application fields of P2P technology for file sharing, audio and video online, search engines, network game, streaming media, scientific computing and collaboration, instant messaging, etc. To give our lives enormous convenient, but also brings a series of problems. P2P business has dominated the Internet business, of 60% to 80% of total become killer broadband Internet applications. Caused the network bandwidth consuming, and even cause congestion. P2P networks ilack the necessary security mechanism, give school network safe protection to opengate door. P2P application of existing technology and management supervision, and the difficulty in educational net P2P users will use within some this vulnerability and the reactionary information spread pornographic, caused a very bad effect. Therefore, we must effectively identify and control P2P traffic.Therefore, effective implementation of the P2P traffic identification and control becoming the current problem to be settled urgently.But the traditional support vector machine algorithm doesn't support incremental learning , as new samples are added , cause training set scale unceasingly to expand, use a lot of computing resources, optimal slowly. In a deep research on support vector distribution characteristics are put forward on the basis of partition weighted incremental support vector machine (SVM) algorithm. The works this paper has done list as follows:Firstly, Introduce P2P network structure and characteristics of analysis and research actuality, introduces the P2P flow of identification of several proposals, analyzes the several proposals in the testing process characteristics and problems.Secondly, Expound the SVM knowledge, puts forward deltas support vector machine improved algorithm of partition weighted incremental support vector machine. This algorithm is effective utilization of generalized KKT conditions and center distance ratio, abandon little impact on subsequent training samples, get boundary support vector sets, and effectively eliminate the training sample. We merger the remained samples, and give weighted processing, solve some sample gravely deviates from belongs to the category, which is not fair for normal distributed samples.Thirdly, Puts forward the partition weighted incremental support vector machine is applied to detect problems of P2P flow ideas. Design and implement the P2P flow test model, and gives partition weighted incremental support vector machine (SVM) method of realization, through rational kernel function parameters, testing time and from two aspects of detecting precision evaluation classifier classification results. Practice proves classifier has very good classification effect.At the end of this paper, we make a conclusion for the design and research of the dissertation. Especially brings forward some suggestion for the paper, and points out the major creativeness future work.