基于显式计算的单播链路时延估计  

Unicast link delay inference based on explicit computation

作　　者：张志勇[1] 潘胜利[1] 胡光岷[1] 

机构地区：[1]电子科技大学宽带光纤传输与通信网技术教育部重点实验室,成都611731

出　　处：《计算机应用研究》2014年第12期3742-3746,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61171091;61201127);电子科大中央高校基本业务费资助项目(E022050205)

摘　　要：为提高基于单播测量的网络链路时延分布估计速度,提出了一种基于显式计算的单播链路时延快速估计方法。由于网络设备多样化导致链路时延特征存在差异,首先根据端到端测量数据为各链路分配不同的离散间隔;然后利用背靠背探测包在网络中引入的二层二叉树结构,通过显式计算推断各链路的时延分布,能明显提高估计速度。仿真结果表明,该方法将链路时延分布的估计时间降低至数百毫秒,同时更准确地捕获链路时延特征。该方法基于显式计算,计算复杂度非常低,因此能够满足实际应用的实时性要求。In order to speed up the inference of link delay distribution in unicast scenario,this paper proposed an explicit computation algorithm. Because of the difference among the link delay characteristics caused by the diversity of network elements,the algorithm designated different bin sizes according to the end-to-end measurements for the links. After that,it took advantage of the two-level binary trees introduced by back-to-back probes to calculate the delay distribution of each link using explicit computations. Simulation results show that this algorithm can reduce the running time of estimating link delay distributions to a few hundreds of milliseconds and capture the link delay characteristics more accurately. This algorithm is computationally simple,as it is based on only explicit computations. Therefore,it satisfies the real-time requirement in practical scenarios.

关 键 词：网络层析成像 单播链路时延估计 可变离散间隔模型 显式计算 

分 类 号：TP393.04[自动化与计算机技术—计算机应用技术]