Analysis And Mining Of Network Flow Connectivity Graphs For Backbone Communication Network

The backbone communication network is the main artery for network information transmission,and it is also an important infrastructure of the Internet.Network flow behavior describes the regulation of dynamic changing during the network operation process,including the characteristics of network flow behaviors and their dynamic changes,network fault and anomalous behaviors.In order to have a better managing and controlling the backbone communication network,it is necessary to analyze and mine the flow behavior of the network in order to obtain the comprehensive and accurate characteristics of the flow behavior of the backbone communication network and its changing characteristics.Thus this could help network managers to enhance their control of the network operation situation,deeply understand the root-causes of changing of network flow behavior,and discover network anomalous events related to the flow behavior among the backbone communication network.With the increasing scale and the emerging of new application services,the backbone communication networks are developing towards high-speed,diversification and big data.As all kinds of events happening in the human social life will also have an impact on network flow behavior.Therefore,traditional network flow behavior analysis methods are facing problems such as the difficulty of information processing,the increasing difficulties for analysis and the increasing cost.It is necessary to have a new thinking and research on the characteristics for the backbone communication network.Since graph modeling methods can accurately capture and present the relationship between data,it has become an efficient method of network flow behavior analysis.Based on the in-depth study of the existing literature,this thesis proposes a new graph modeling method ——Network Flow Connectivity Graph(NFCG).Combining with the latest research achievements in the fields of complex networks,big data mining and signal processing,this thesis studies the construction of NFCG graph model,the extraction and analysis of graph characteristic features,the analysis of subgraph patterns and the analysis of graph dynamic evolution.The innovative achievements in this thesis are listed as follows:1.Firstly,there have problems of simple construction and incomplete information embedded in existing graph model methods.By drawing on the idea of knowledge graph,this thesis proposes a method for constructing NFCG,which can assign different attributes and ranking-levels to nodes and different weights to edges according to the attribute information of network flows.According to different research purposes,various filtering rules for nodes and edges have been developed for extracting the core relationships of NFCG,while we also could use edge definition rule to generating various NFCGs which have different physical representation.Compared with the existing methods,NFCG constructing method is flexible and contains richer flow connected information.In a word,the construction of NFCG is the basis of subsequent work for network flow connectivity behavior analysis and mining.2.In order to break through the limitations of the existing feature analysis methods for graph model,this thesis starts from the global graph level and draws lessons from the research ideas of complex network feature analysis.Regarding NFCG as an enhanced complex network structure,we could use different eigenvalues to describe NFCG,and extract a variety of features suitable for different scales.By analyzing the characteristics of graph features of different application behaviors and comparing the differences of graph features between normal and abnormal network flow behaviors,the proposed method in this thesis can achieve accurate classification for network application flow behavior and identification of P2 P applications.Furthermore,we can also effectively mine the abnormal flow behavior which is difficult to achieve by existing methods.3.The changes of flow connectivity behaviors are distinctive for various network events,and they can be reflected by subgraph pattern.From the point of view of subgraph pattern analysis and associated subgraph pattern mining,this thesis proposes two methods to apply NFCG to network event analysis and host identification of Internet data center: 1)According to the different characteristics of network events,a subgraph pattern matching method is proposed,which extracts effectively from specific subgraph structure,edge similarity and frequent subgraph.The subgraph pattern associated with network events is helpful to mine the root cause of network flow behavior;2)Association subgraph pattern refers to the subgraph pattern with strong correlation among nodes.A mining method of association subgraph pattern is proposed and applied to the analysis and mining of flow connected relationship among in Internet data center.This method can effectively identify IDC hosts and at the same time according to the behavior similarities of nodes,it can help a lot to infer the sevices' types for an unknown IDC.4.Aiming at the characteristics that network abnormal events usually cause network epidemic behavior and abnormal changes of its evolutionary characteristics,this thesis proposes a novel method for network anomalous events detection and identification based on evolutionary analysis from the perspective of graph feature evolution and subgraph pattern evolution of NFCG: 1)An abnormal event identification method based on graph evolutionary characteristics is proposed.Firstly,the flow connected relationship is extracted.Then various evolutionary characteristic features of graph sequence are employed,so that we can use outlier-detection method to find the abnormal points in evolutionary characteristic features of graph sequence.The simulation results show that the method can detect network events effectively,and matching with the characteristic features of anomalous flow behavior can also help to identify the types of network anomalous events;2)Taking network motifs(a small induced subgraph structure)as the research object,the mining method of frequent motifs in the evolution process has been studied,and an algorithm based on cascading motif discovery is proposed and applied.For network worm detection,the advantage of this proposed method is that it can accurately describe the evolution process of network worm flow connected relationship by using cascading motif structure.Compared with the existing methods,this method can detect more network worms accurately and efficiently,and does not need the cooperation of prior knowledge about network worms.