一种基于端到端的链路报文丢失性能推测方法

A Method for Link Loss Inference Based on End-to-End Measurement

网络链路性能参数对网络性能评价和管理非常重要,为了克服目前网络链路性能参数的统计推测方法中存在的过低估计问题,提出了一种逻辑链路报文丢失的累计生成函数(CumulantGenerating Function,CGF)的推测方法。根据端到端报文丢失的CGF值,利用最小二乘法可推测每条逻辑链路报文丢失CGF值。仿真试验证明,推测所得的CGF值与理论计算值很接近,算法可以很好地解决链路报文丢失性能的推测。同时,根据推测结果,利用Chernoff边界定理还可以识别报文丢失严重的链路。

Aim. Existing algorithms for inferring network performance have an overfitting problem. We propose what we believe to be a better algorithm to overcome this problem. In the full paper, we explain in some detail the algorithm we propose, called by us CGF （Cumulant Generating Function） algorithm. In this abstract, we just add some pertinent remarks to listing the four topics of explanation. The first topic is problem formulation. In the first topic, we propose the network model for the inference of link performance. The second topic is the inference of the link packet CGF from the end-to-end path packet loss CGF. In the second topic, under the assumptions that the link losses are mutually independent, we elaborate a bias corrected link loss cumulant generating function algorithm. The third topic is simulation results. In the third topic, we compared the sampled CGF value of link loss rate with estimated CGF value of link loss rate. The experimental results, given in two figures （4 and 5）, show that our proposed algorithm is accurate and effective. The fourth topic is lossy link identification. In the fourth topic, based on the estimated CGF value, we can identify the lossy link in the network using Chernoff Bounds theorem with high reliability, and the results are shown in table 1.