基于Manhattan距离的网络拓扑推测方法研究

Network Topology Identification Based on Manhattan Distance

网络拓扑推测是网络断层扫描研究内容之一,是推测网络内部链路性能的前提条件。目前的网络拓扑推测方法主要是基于网络性能参数的极大似然估计方法,随着网络规模的增加,计算量相对较大,还有些需要先验知识,从而影响在实际网络中的应用。为了克服这些问题,作者提出了一种新的网络拓扑推测方法,通过直接计算节点间测量数据的Manhattan距离,对节点进行分组,从而推测网络的逻辑拓扑结构。该方法计算简单且准确推测的概率收敛速度快,在实际网络环境中有应用价值。

Network Identification, one of the studies of the network tomography, is the proposition of the network link-level performance inference. The present methods rely either on the network performance or on the posterior distribution, and the time spent on the identification increase as the size of the network, which may restrict the technique to be used in practice. To overcome the above problems, we propose a fast approach to identify the logical network topology in this paper. Compared with the previous methods, the proposed one only needs to calculate the Manhattan distance based on the measurements to identify the network topology, which saves more time than the present ones, and the time spent on the identification do not increase sharply as the size of the network. From the simulation study, we find that the fraction of correctly classified trees fast converge, and accurately identify the trees even under the conditions that just hundreds of probe packets are injected. So the proposed method is very promising in the real network.