一种单播网络拓扑推断方法

现有的基于三元分组列车的网络拓扑推断方法,仅依据单一参数(时延或时延抖动)进行网络拓扑的推测,其推断准确度受到网络负载的影响,在网络负载较重时不能准确推断出网络拓扑.针对上述的问题,提出一种基于三元分组列车测量拓扑结构的方法,利用叶节点的层析信息将叶节点进行聚类,依据时延抖动和丢包率两个参数计算节点间的相关性,由底向上构造网络拓扑树.经过NS2仿真环境下实验验证,该方法有效减少了探测包的发送量,提高了推断的准确度,且网络负载对推断准确性影响较小.

A Topology Inference Method of Unicast Network

Since most of existing network topology inference methods based on the ＂sandwich＂packet sequences only use one parameter（time-delay or delay-jitter）to speculate the network topology,their inference accuracy is easily affected by network load,and it can not accurately infer topology when network load is heavy.In order to solve the above-described problems,a new method is proposed based on＂sandwich＂packet sequences to infer network topology.The method clusters the leaf nodes by its heigh,calculates the correlation between two nodes according two parameters（delay-jitter and packet loss rate）and constructs network topology tree from the bottom to the top.The simulation in NS2 proved that the new method can reduce the amount of probe packets,raise the accuracy and has less influence on the accuracy of network load.