The Research Of Network Topology Inference Accuracy Based On Unicast

With Internet technology deep into every sector of social life such as the financial sector,communications services industry, coal industry, the oil industry, food distribution industries,the relationship between people’s daily lives and the Internet increasingly close. At the sametime, the network requirements for real-time, accurate and reliable transport services are alsoincreasing, especially in network transmission latency and network security. As thefoundation of network management and monitoring, network measurement becomesincreasingly important. Topology inference is an important aspect of network measurement,which is used to protect all kinds of Internet activities carried on smoothly. The networktomography (NT) is an end-to-end measurement technology by sending probe packets from asource node to a particular destination node, and getting various statistical parametersinformation (packet loss rate, delay, jitter and link bandwidth,etc.) of the probe packetarrives their destination node, and according those information to infer sibling relationshipbetween nodes. The NT technology effectively overcomes the shortcomings of traditionalnetwork measurement and already becomes the most effective technique of inferring thenetwork topology.According to the different ways of collecting network link’s parameters, the topologymeasurement techniques based on network tomography can be divided into two categories:one is active measurement, another is passive measurement. The active measurement can alsobe classified into two types of measurement based on unicast and measurement based onmulticast. Since the devices character of supports multicast functions or not, is no directlyrelated to network security. And there existing some network devices do not supportmulticast functions, which limited the application range of multicast measurement method.Whereas the measurement methods based on unicast have lowly demands for networkequipment itself, thus unicast methods have more widely range of applications. This studywill be promoting on measurement of unicast network topology. The main contents are asfollows:First, in this paper, it’s a summary exposition of the network topology measurementtechnology’s background, significance and its development status around home and abroad.And then we introduced traditional network measurement techniques and the newmeasurements method which is based on NT technology. Following, we compared thetraditional measurement method with the method based on NT, and analyzed advantages anddisadvantages of the traditional methods.Second, it is a description about two end-to-end measurements way, include activemeasurement method and passive measurement method, and the two sorts of active network measurement method which are unicast-based and multicast-based measurement. Then thispaper described some probing packet’s patterns of unicast measurement, and analyzed boththeir advantages and disadvantages.Then, this paper discussed the existing three-packet train sequence measurement modeland two measurement methods based on the three-packet train which uses delay or delay-jitterto infer network topology. Basis of the existing three-packet train sequence measurementmethods which are based on single parameter (delay or delay jitter), we propose a newthree-packet train sequence measurement method which uses double-parameters to infer thetarget network topology, systematically illustrate the whole measurement process. And weexecuted simulation experiment on the process of sending probing packets under NS-2environment, collected relevant information data of those packets at the same time. Theexperiments executed under different load conditions of the network, and compared theinference accuracy between the existing method and the new method. The results demonstratethat the new method has the effectiveness and accuracy.Finally, this paper discussed the situation of applying the improved method to infer thenew topology when there are new nodes added in topology-known network, and verified theaccuracy in NS-2simulation environment. This method brought the height of leaf nodes in theprocess of inferring network topology, so this method effectively reduced the amount ofprobing packets sending when inferring the topology with complex nodes level.