Measurement-based Network Management Technology

After more than a century of development, the network has brought dramatic changes to people's lives, but we should see that network management is faced a lot of challenges causing by the highly heterogeneity and openness of network and diversity of network services. Network measurement technique which has become a hot topic of current research grasp the actual the network state through real network measurements to achieve an objective evaluation of the network.In this thesis, the requirements of network measurement, the importance of network measurement in network management, the practical significance of network measurement are analyzed first. Then, current research situation, research advances and characteristics of research institutions in network measurement, especially in performance measurement are summarized and reviewed. Next, research methods of network measurement, technologies related to performance measurement and some key issues in network measurement are deeply studied. Among them, network performance measurement is elaborated in detail on the base of passive measurement technology.To reduce the redundancy of the collected flow data, data fusion algorithm in feature level is used to conduct performance data fusion. For the flow data collected by measuring hosts, feature extraction is done according to performance index at all levels, and then fuzzy neural network is adopted to fuse these feature vector layer by layer, in this FNN, BP algorithm is employed in off-line training in which error is back propagated to adjust parameters and accelerate convergence, and particle swarm algorithm based on an adaptive inertial factor dynamic adjustment strategy is used to optimize fusion performance and reduce the time complexity and space complexity.On the basis of above fusion algorithm, aiming at the ambiguity of network performance parameters, knowledge of fuzzy theory is introduced. the corresponding secondary indicators are refined from delay, bandwidth, packet loss rate and other key performance indicators, then a network performance evaluation model based on fuzzy theory is proposed. Specific data shows the model can evaluate the fused performance data reasonably and effectively. In practice, the design and implement of the network traffic measurement system is detailed analyzed from the aspect of software architecture and software engineering respectively: the deployment framework, architecture, and system design, the design of each module are described; important data structures, interfaces, error handling and operation are narrated; system database, main class diagrams, use case diagrams and system dynamic model are detailed design, especially, the design of system core module and its algorithm are highlighted; ultimately, for the major network performance indicators, actual performance testing on the prototype system is conducted with the help of a good experimental environment and equipment advantage of CERNET Network Center in Southwest China.