Reseach On Topology Tomography Based On Network Traffic Behavior Characteristics

With the popularity and development of the Internet,the network topology presents a trend of diversification and complexity.Accurate and real-time acquisition of the network topology is of great significance for strengthening network supervision,improving network service quality,alocating network resources,and optimizing network structure design.Different from traditional network detection methods,network tomography technology can perform network topology inference through end-to-end measurements without the cooperation of internal network elements.The current network tomography methods general y use the end-to-end delay and loss as the input of topology inference,but the end-to-end parameters of a single characteristic cannot ful y reflect the topology information of the network,and are only suitable for network environments under specific loads.Moreover,the data quality of end-to-end measurements is unstable,and is prone to interference from data noise.All these issues affect the accuracy of network topology inference.In order to utilize more comprehensive network traffic characteristic information to improve the performance of topology inference,this thesis carries out the research on topology tomography based on network traffic behavior characteristics,and the main contributions are summarized as follows:(1)A topology inference method based on multi type network traffic characteristics is proposed.This method extracts and processes traffic data in the network through passive traffic monitoring to obtain multiple types of characteristic information that can reflect the topology of the network(such as packet round-trip time,packet time interval,flow rate,and other network traffic behavior characteristics).Then,the multiple types of network traffic characteristics are converted into metric parameters for the similarity of destination nodes,and the destination nodes are aggregated from bottom to top using hierarchical clustering.Compared with the topology inference method that only uses a single end-to-end parameter,the method proposed in this thesis can utilize more comprehensive network traffic characteristics,and improve the accuracy of topology inference without adding extra network loads.(2)A network topology inference method based on data quality adaptive optimization is proposed.This method can adaptively select the time period with strong correlation of detection data and a set of destination nodes in this time period through data smoothing filtering,node similarity analysis,and destination node selection,and then restore the substructure containing only part of the destination nodes.Finally,substructure fusion is performed based on the new node similarity metric parameters to obtain the complete topology.This method reduces the influence of unstable quality of probe data and improves the accuracy of topology inference.This thesis conducts comparative experiments in the actual network environment and on the NS3 simulation platform,and proposes the accuracy evaluation index based on the perspective of internal node matching.The experimental results show that the proposed method improves the effect of topology inference in terms of accuracy and stability.