The Analysis And Modeling On Internet Topology's Community Structure

Since the discover of the power-law relationship in Internet topology, research on the complexity of Internet topology brought more and more people's concern, the analysis and modeling on Internet topology is a popular topic. There is great achievement in this area in recent years and many unknown topology characteristics have been revealed in recent years. However, there is still some blank in this area, for example, community structure is a common property in many real networks, but there isn't any research on the community structure of Internet topology, or related modeling method.In this work, the community property of Internet topology is analyzed, on the basis of the mass topology data provided by CAIDA. First, a basic analysis on community property for simple network models using modularity curve shows that random network has a very strange behaviors compared with other simple network models. And the analysis on real Internet topology shows that Internet AS-level topology has a modularity around 0.40, which indicates that Internet is also a network with community structure, while at the same time, most of the popular Internet models'modularity is less than 0.30, indicating their inconsistence with the real Internet.The analysis on the origin of community structure of Internet topology shows that most ASes in the same community belongs to the same country or adjacent countries, which means that geographical consideration is an important cause of the community structure in Internet topology. The influence of AS classes on this property is relatively weak. A further analysis on Internet country-level topology shows that the several communities correspond to the several continents in the world, which further strength our conclusion.Connection cost is important aspect which should be considered when deploying real networks. Because of the consideration to reduce the cost, the probability to establish connections between adjacent areas is relatively large. A new model which has taken geographical location into consideration is introduced to model the evolution of Internet topology. In the new model, which is called GeoPFP model, there is more preference for new nodes to connect with nearby nodes. And the experiment show that the new model is in consistent with the real Internet in most graph properties, and has a similar modularity as Internet, which further proved our conclusion.Compared with current Internet models, GeoPFP model has obvious advantages and can be further used in topology-related researches, such as being used to develop a more precise topology generator. Also, it is significant for the development of next-generation Internet and better routing protocols.