摘要： 现有时延层析算法大多考虑离散时延模式，但算法效率比较低。为此，提出一种连续时延估计算法，假定链路时延为某参数的函数分布，根据多播特征并基于端到端数据的矩，利用非线性最小二乘法估计链路时延分布函数的参数，并在每步迭代中用一维牛顿搜索确定最优步长，达到快速收敛。应用Matlab和NS2仿真软件得到的数据表明，该算法所需的存储量少，算法简单且效率较高。

关键词: 网络测量, 网络层析, 时延估计, 多播, 非线性最小二乘法

Abstract: The up-to-date algorithms on delay tomography are based on the discrete delay mode, but it is low efficiency in this mode. This paper proposes a continuous delay mode, whose delay is distributed on a function with some parameters. According to the rule of multicast and based on end-to-end moments, the parameters can be estimated using nonlinear least squares and the best step is chosen by the one-dimension Newton search in iteration. Experimental results on Matlab and NS2 simulation show that the algorithm needs little storage and is simple with high efficiency.

Key words: network measurement, network tomography, delay estimation, multicast, nonlinear least square method

中图分类号: 

• TP301.6