Virtual Test Network Parameters Estimation Based On Tomography

Virtual test technology has been widely appplied in defence industry and used as an important method to evaluate the performanc of weapon system. Virtual test system is created by integrating three types of test resources: live equipment, virtual model, and constructive model according to the test task requirements. Generally, virtual test network is built by specific network which is located in multiple geographically distributed ranges. The data exchanges of virtual test system readily generates heavy load, which causes network congestion and transmission performanc degradation, thereby results in time-space inconsistency and affects the entire performance of virtual test system. By obtaining the internal network parameters accurately in real time, and adjusting the system with dynamic optimization and re-deployment according to the variation of internal network parameters, the performance of virtual test network can be improved quickly, and thereby the entire performance of the virtual test system can be enhanced.As virtual test network tends to be non-cooperative, the traditional direct measurement methods, which rely on the cooperation of internal nodes, are difficult to satisfy the requirements of the virtual test network measurement. Recently, network tomography, which is based on end-to-end measurements, is proposed to infer the internal network parameters without the cooperaton of internal nodes. Network tomograhy provides a new method to solve problems of virtual test network measurements, but existing methods still has disadvantages of high computational complexity and low measurement efficiency. In virtual test system, the efficiency of network measurement is important due to the need of adjusting the deployment scheme in real time. The focus of this thesis is on obtaining the internal network parameters for optimizing virtual test system. In order to improve the measurement accuracy, and reduce the network load caused by the measurement process and computational complexity, some methods are proposed in the fields of network parameters estimation with network tomography. The main contents are as follows:For existing topology inference methods based on network tomography, background traffic has a great impact on the accuracy. In virtual test system, the frequently data exchange process is prone to cause heavy network load, which affects the accuracy. In addition, the members of virtual test system joining and leaving dynamically according test schemes. When a new destination node joins, the existing algorithms for topology with a changing set of nodes exploit exhaustive or hierarchical search methods to update the topology, which leads to low efficiency. In order to improve the accuracy and efficiency, a novel topology inference algorithm based on probe packets recombination and common path matching is proposed. In this algorithm, in order to improve the estimating precision of similarity metric, two small packets of sandwich probes are recombined in accordance with cross-traffic effects, and the similarity metric is estimated according to the new recombined sandwich probes. The new joined nodes are directly added into existing topology by matching the length of common path. By utilizing the information of TTL(Time-To-Live) hop count to select match path, the efficiency of topology inference is improved. Experimental results show that this algorithm can effectively improve the accuracy and efficiency of topology inference.In order to improve the efficiency of link loss inference algorithm of single source network, a fast loss inference algorithm based on subtree loss pattern is proposed. The iterations of the algorithm are reduced by obtaining a more appropriate initialization of loss rates. According to the outcomes of end-to-end measurements, this algorithm partitions the network topology into one area of which the transmission state is determinate, and several areas of which the transmission state is indeterminate. By decomposing all the indeterminate areas, a subtree loss pattern database is constructed. The loss rate is calculated based on the loss pattern. Through the reducing of the redundancy decomposition process, it can speed up the process of the inference of the link loss rate. Experimental results show that the algorithm can improve the efficiency without loss of the accuracy.For link loss inference algorithm of multiple source network, which modelled by a set of high-order polynomials, the order of polynomials is determined by the degrees of internal nodes. And the order of polynomials determines the computational complexity. In order to improve the efficiency of the link loss inference algorithm of multiple source network, a novel multiple source network link loss inference algorithm based on topology transforming. The multiple source topology is transformed according to the relationship of the order of polynomials and the number of branches of internal nodes. And then by using the transformed topology, the order of polynomials is reduced. In order to minimize the estimation error, the algorithm uses minimum mean square error criterion to select the optimal scheme of topology transforming. Experimental results show that the algorithm can significantly reduce the time of link loss rate inference with identical accuracy.