基于TCP被动测量的单播网络丢包层析

在大规模网络中，网络层析通过端到端测量推测网络内部性能，能有效减少网络内部设备之间的协作，因 此被广泛应用于网络性能诊断。大多数网络丢包层析都采用主动发送探针的方法，无法准确反映实际网络的性 能。为了避免上述局限性，介绍了一种基于TCP监测的被动测量网络丢包层析方法，通过端到端TCP数据流的采 样，采用非因果条件概率构造似然函数，并通过EM 算法求解最大似然估计推测网络内部丢包率。理论分析表明 该方法具有一定的可行性。

Unicast network loss tomography based on TCP passive measurement

Network tomography is a new technique for the internal behavior of large-scale networks based on end-toend measurements. It minimizes the cooperation from network elements and has been widely used in characterizing the performance of a large-scale network. Most network loss tomography methods utilized active probing, but the performance obtained from probes is not equal to the reflection of actual network. In this paper, an alternative net- work loss tomography based on TCP monitoring and anti-causal conditional probability is introduced to avoid these limitations. And EM algorithm is used to compute maximum likelihood estimates of internal loss rates. Theoretical analysis demonstrates the feasibility of this method.