Technology Research On P2P Flow Application Layerreal-time Detection

B/S or C/S mode, is currently the most commonly used network application mode. Itis based on a server as the center, all kinds of information by the server center passed to allclient node. Information transmission mode, generally can be concentrated and stored to theserver, and can be downloaded separately, or the information is the central serverproprietary software processing, they can transfer and flow between network users.P2P（Peer-to-Peer）is called the " point"," equivalence" technology, is an emerging networktechnology, it is dependent on the network computing power and bandwidth of each noderesources, than the traditional C/S model has great superiority，including distribution,scalability, extensibility and high performance-price ratio. It is in network television, filesharing, distributed computing, network security, online communication and evenenterprise computing and electronic commerce and other application areas have a very widerange of applications.P2P strategy broke the traditional Client/Server network architecture,which brought great conveniences to users, while bringing some questions such as thesecurity and the resources invaded.In order to solve these questions and make sure the P2P net work can be better usedby the clients; we must control the P2P application. At first, we should distinguish the P2Ptraffic from the all kinds of the network traffic. However the tradition model of P2P trafficeidentification strategy takes some effects on the identification, the strategy was toomonotonous and the shortcoming was obvious. Especially it can not satisfy therequirements of real-time identification, so we should develop new strategy to make upthese shortcomings. When the strategy of machine-learning was brought into the field oftraffic identification, it is turn for the better. Although the strategy was able to adjust itselfto network changes, the accuracy of the distinguish was precise, yet the expense of trainingwas too expensive and the speed of the distinguish was too slow, thus it still tooinefficiency to meet the requirements of real-time. In addition, with the use of the P2Pplatform such as the P2P internet TV live, we put forward higher demand to supervise theP2P effectively. Not only can we distinguish whether it is P2P traffic, but also distinguishthe specific application platform, in this way we can control it in a better way.Innovation of this paper includes three aspects, one is the design of a kind of entropyoptimization improved support vector machine model. The model from the reduced samplespace scale angle, eliminate redundant vector, retention of decision functions play a decisive role in the support vector, thus shortening the training of support vector machinetime cost. Two is for the P2P traffic detection of three kinds of different levels of demand,through the theoretical analysis and the experimental summary of the way, put forward forthe identification of three kinds of flow characteristic vector. Vector in a variety of features,is a real-time, thus realizes to the P2P real-time traffic detection requirements. The three isto design and implement a prototype system, online and on campus were deployed to test,prove the availability of the system.According to the above, the thesis from the specific question put forward entropy toimprove the SVM. By reducing the sample sets scale to seek the methods to support thetraining SV, put forward a kind of improved entropy optimization of support vectormachine method, it improved the training efficiency of the SVM and decreased thespace-time expense, thus it could meet the real-time demand. The optimization of thedetection method based on P2P in this paper, the characteristics of the transport layer arestudied, especially for P2P and P2P, P2P document flow and P2P multicast video streamand P2P stream video multicast application level to distinguish between correspondingfeatures, design two dimensional feature vectors, a ten dimensional characteristics vector isused to detect. This paper theoretically analysis at the same time, design and developmentof the flow detection system in campus network, and the deployment of the experimentaldetection. By analyzing the results of experiments, proves that the system is capable of P2Pflow in real time detection, and can adapt to network changes, and on the P2P videomulticast flow of application level differentiate, meet the administrators online real-timemanagement and control requirements, verify the feasibility of the method. At the sametime, this paper combining with the practical study, in the future is predicted and prospect,and the combination of personal reflections, the prototype system puts forward some adviceon further improvement. The prototype system only on P2P flow identification were studied,and not on the management part of the design, in the future study and research will furtherrefine.