基于主动探测的IP网路径及链路丢包率推理算法

发布时间：2018-04-24 04:10

本文选题：网络层析成像 + 性能管理　；参考：《北京邮电大学》2014年硕士论文

【摘要】：随着互联网的迅速发展,人们对网络服务的依赖程度不断提高,对网络的性能也提出了更高的要求。网络性能测量是进行网络拓扑、流量、行为建模分析的基础和验证手段。如何用较低的代价,准确地测量得到整个网络的运行状况是网络测量研究中的主要内容。在IP网传递信息的过程中,网络端到端路径丢包率和网络链路丢包率能够反映网络的运行状况和拥塞程度,是网络性能的重要指标。因此如何快速准确地得到路径与链路丢包率成为众多研究者研究的热点。本文首先提出了一种基于网络层析成像的路径丢包率推理方法,可通过测量r条分布均匀的独立路径较准确地推理出剩余路径的丢包率。新方法利用先前探测返回的结果自适应的选择下一个探测路径,对非拥塞的路径的删减使得路由矩阵规模变小,从而大大降低了独立路径个数。并且该方法不需要求解链路丢包率的值即可得到所有剩余路径丢包率的值,从而大大降低了算法复杂度。实验表明新方法在保证推理准确度的条件下,筛选出的独立路径条数较已有经典方法减少了5%-14%。由于现有的网络链路丢包率推理算法端到端探测比较困难、算法复杂度比较高,为了解决上述问题,本文还提出一种性能比较高的、新的链路丢包率推理算法。新算法首先建立代数模型,得到部分可确定链路的通过率,然后将可确定链路从方程中移除,之后将链路划分为若干独立子集,最后将每个子集的方程组求解问题转化为非线性规划的最优化问题,从而在运算时间尽量短、网络探测尽量少的情况下推理求得比较准确的数据。我们通过仿真实验与LIA算法以及LIABLI算法进行了对比、和分析,尽管新算法仅用了2%的探测,但是诊断准确度、推理准确度以及运行时间明显得到改善。
[Abstract]:With the rapid development of the Internet, people rely on network services constantly, and put forward higher requirements for the performance of the network. Network performance measurement is the basis and verification method for modeling and analyzing network topology, traffic and behavior. How to accurately measure the operation of the whole network at lower cost is the main content of the network measurement research. In the process of transmitting information in IP network, the network end-to-end path packet loss rate and network link loss rate can reflect the network performance and congestion degree, which is an important index of network performance. Therefore, how to quickly and accurately obtain the path and link packet loss rate has become a hot topic for many researchers. In this paper, a path packet loss rate reasoning method based on network tomography is proposed, which can accurately infer the packet loss rate of the remaining path by measuring the independent path with uniform r distribution. The new method adaptively selects the next detection path using the results of the previous detection returns, and the deletion of the non-congested path makes the routing matrix smaller, thus greatly reducing the number of independent paths. Moreover, the method can get the value of all the remaining paths without solving the value of the link packet loss rate, which greatly reduces the complexity of the algorithm. Experiments show that the new method reduces the number of independent path bars by 5-14 compared with the classical method under the condition that the reasoning accuracy is guaranteed. Due to the difficulty of end-to-end detection and the high complexity of the existing network link packet loss rate reasoning algorithms, in order to solve the above problems, a new link packet loss rate reasoning algorithm with high performance is proposed in this paper. The new algorithm first establishes an algebraic model to obtain the pass rate of a partially determinable link, then removes the determinable link from the equation, then divides the link into several independent subsets. Finally, the problem of solving the equations of each subset is transformed into an optimization problem of nonlinear programming, so that the more accurate data can be obtained by reasoning under the condition that the computing time is as short as possible and the network detection is as little as possible. The simulation results are compared with LIA algorithm and LIABLI algorithm. Although the new algorithm uses only 2% detection, the diagnostic accuracy, reasoning accuracy and running time are improved obviously.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【参考文献】