【作者】 杨京礼；

【导师】 姜守达；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2013， 博士

【摘要】 虚拟试验技术凭借其低成本、低风险等优势广泛应用于武器装备研制生产过程中，成为军工试验与测试领域的发展趋势。在基于虚拟试验技术构建的系统中，时空一致性决定了系统的公平性和正确性，是试验结果真实有效的保证。为提高虚拟试验过程的时空一致性，需要实时获取网络当前的性能参数和拓扑结构信息，以此为依据及时调整实体在各网络节点的部署，优化虚拟试验系统的整体性能。因此，虚拟试验网络测量是改善虚拟试验系统整体性能的基础，而测量方法的实时性是保证虚拟试验真实性和有效性的前提条件。随着虚拟试验技术的发展，广域、跨系统边界和虚实结合的试验模式使得虚拟试验网络具有异构化和非协作等特点，传统的直接测量模式难以满足虚拟试验网络测量的需求。近年来，国际上多个研究机构提出的网络层析成像技术以端到端模式解决网络的非协作测量问题，成为网络测量领域的研究热点。目前，网络层析成像技术的研究主要针对互联网测量而开展，所关注的重点在于提高大规模网络性能参数的测量精度，现有的方法难以满足虚拟试验网络测量的实时性需求。针对上述问题，本文在网络层析成像技术框架下，从虚拟试验网络性能优化相关的参数（丢包率和时延）测量和拓扑结构估计两方面进行研究，力求提高测量效率（实时性）和测量精度，降低测量过程造成的网络负载，具体内容如下：在网络层析成像技术框架下，端到端测量路径的数量决定了测量的效率和网络负载情况，而完成端到端测量所需要的最少路径数量等于路由矩阵的秩。针对现有的链路丢包率测量方法存在的测量效率较低、测量过程易造成较大网络负载和测量精度不高等问题，本文提出一种基于最小覆盖集的链路丢包率测量方法。该方法在端到端测量之前首先通过最小覆盖集测量寻找网络中的正常链路，化简路由矩阵，缩小路由矩阵的秩，减少端到端测量所需要的路径数量；随后采用线性方程组和Gibbs采样相结合的方法求解网络内部链路的丢包率，改善由于采用高斯模型描述网路链路报文传输概率导致测量精度降低的问题。仿真实验结果表明，该方法可以有效提高链路丢包率的测量效率和测量精度，降低测量过程造成的网络负载。在离散时延模型下，端到端路径时延分解后的基本计算单元数量越少，链路时延分布测量方法的计算速度越快。针对现有的基于离散时延模型的网络链路时延分布测量效率较低，且在测量精度和计算时间方面相互制约的问题，本文提出一种基于层次分解的链路时延分布快速测量方法。该方法按照网络拓扑结构的层次对端到端路径时延进行子树分解，以子树时延作为基本计算单元进行链路时延分布的计算。由于子树时延是子树所包含链路的时延组合，因此将端到端路径时延分解到子树时延后的基本计算单元数量显著少于分解到链路时延后的基本计算单元数量，使得该方法的计算速度得到显著提升。仿真实验结果表明，该方法可以在保持现有EM算法测量精度的条件下，有效缩短测量时间，提高链路时延分布的测量效率。通过对现有的深度优先搜索下的网络拓扑结构估计方法的分析，发现在对目的节点进行深度优先搜索排序的过程中，参考节点的选取决定了拓扑结构估计的效率。针对现有的单播测量模式下的网络拓扑结构估计方法效率和准确性较低的问题，本文提出一种高效的单播网络自适应拓扑结构估计方法。该方法利用探测包中的TTL信息选取参考节点对目的节点进行深度优先搜索排序，减少单播测量中所需的背靠背包对数量，提高拓扑结构估计的效率；在拓扑结构计算过程中，根据链路加性特征量的估计值实时修正判定阈值，使其对初始判定阈值的选择具有自适应性，解决网络内部链路参数未知时的判定阈值选择问题。仿真实验结果表明，该方法具备较高的拓扑结构估计效率，并且在网络内部链路性能参数未知的情况下仍然具有较高的拓扑结构估计准确率。更多还原

【Abstract】 Virtual test technology has been widely applied in the process of weapondevelopment because of the advantages of low cost and low risk, and has becomethe trend of military test. In the system constructed by virtual test technology, thetime-space consistency is the key issue of the correctness and fairness of the system,and it also assures that the test results are real and effective. In order to ensure theauthenticity of the virtual test process, it is necessary to obtain the networkperformance parameters and topology information in real time, and adjust thedeployment of the entities in each network node. Therefore, the measurement ofvirtual test network performance parameters and topologies are the basis ofimproving the performance of the virtual test system, and the real-time measurementmethods are the prerequisites of authenticity and validity for the virtual test.With the development of virtual test technology, virtual test network tends to beheterogeneous and non-cooperative because of the test mode of wide areas, acrosssystem boundaries and virtual-actual combination. It is difficult to meet the needs ofthe virtual test network measurement by traditional direct measurement mode.Recently, network tomography is proposed to measure the internal networkparameters without the cooperaton of the internal nodes, which has become one ofthe focused research technologies in the field. Currently, the research of networktomography is focused on the measurement accuracy and implemented on Internet.It is difficult to meet the real-time requirements of the vitual test networkmeasurement using existing methods. In order to improve the measurementefficiency and accuracy, and reduce the network load caused by the measurementprocess, some methods are proposed in the fields of network parameters (loss rateand delay) measurement and network topology inference with network tomography.The main contents are as follows:In the field of network tomography, the number of the measurement paths is atleast equal to the rank of the routing matrix, which determines the efficiency of themeasurement and the network load caused by the measurement process. In order toimprove the accuracy and efficiency of the inference method for the link loss rate, anovel link loss rate inference method is proposed based on minimal cover set. Theproposed method decreases the number of the end-to-end paths by reducing the rankof the routing matrix caused by the minimal cover set measurements. Apart fromthis, in order to solve the accuracy problem caused by using Gauss model, theproposed method calculates the link loss rate by the combination of solving linearequations and Gibbs sample. Simulation results show that the proposed method can obtain a higher accuracy with less end-to-end measurement paths.In the discrete delay model, the fewer number of basic computational unit forend-to-end path delay decomposition, the faster link delay distribution calculationthere will be. In order to speed up the inference methods of the link delaydistribution based on discrete delay model, and resolve the mutual restraint inaccuracy and time of calculate, a fast delay inference method based on hierarchydecomposition is proposed. This method decomposes the end-to-end path delay intosubtree units by the levels of the topology, and calculates the link delay distributionbased on those subtree units. Because the number of basic computational unit isfewer than that decomposed into link delay units, it can speed up the process of theinference of the link delay distribution. Simulation results show that the proposedmethod can improve the efficiency of the link delay distribution without losing theaccuracy of the EM algorithm.In order to improve the accuracy and efficiency of the topology inferencealgorithm for unicast network, an efficient and adaptive topology inference methodis proposed. With the information of TTL hop count, this method reduces thenumber of the probe pairs needed in the process of bisection Depth-First SearchOrdering, and improves the efficiency of the topology inference. In addition,through the analysis of the principle of the Depth-First Search topology inferencealgorithm, a sufficient condition for the algorithm to return the correct networktopology is given. Based on this condition, an adaptive threshold selection method isproposed, which can improve the accuracy of the topology inference when thenetwork link parameters are unknown. Simulation results show that the proposedmethod can obtain a higher accuracy and efficiency.更多还原