Topological Characteristic Research And Modeling For AS-Level Internet

The rapid development of Internet has brought a series of problems, such as routing, resource reservation and network management, and other issues. As experimenting on the Internet is very expensive, and also for some commercial factors, the researchers cannot experiment directly on the Internet, but often on network topology models. Internet topology modeling provides an abstract model structure for the network analysis, and makes many of problems is easy to be analyzed. Therefore, the research of modeling the Internet topology has been an opening topic.The Internet topology modeling started from the random model to the hierarchical model. Then it developed to a more realistic one, the scale-free network model. But now, researchers have not proposed a unified theory to explain the evolution of the Internet. They establish evolution models in accordance with their own analysis, and then evaluate their models according many parameters, to make their models approximate to the actual Internet topology, as well as explain evolution of Internet rationally as can as possible.This paper first introduces four basic statistical characteristics of the Internet topology: scale-free, small world, clustering and disassortative mixing. And those characteristics have been tested on current Internet in this paper.Secondly, this paper analyzes many dominical topology generators at present, including BA, AB, Inet, PLRG and GLP model, and then uses some authoritative metrics to evaluate them. Results show that topological maps generated by these models can exhibit some topological features of the Internet, but there are also certain deficiencies, especially the clustering all are inferior to Internet.Thirdly, this paper analyzes the limitations of the evolution algorithms used by those models. Inet and PLRG model are static, and their aim is to generate an approximate map with power law characteristic. Static models cannot explain the evolution process of the Internet. Dynamic models, such as BA, AB and GLP model, establish evolution mechanisms to simulate the development of the Internet. But those models made uniform growth assumptions, which didn't meet the exponential growth of the Internet, and their attachment mechanisms are simplex.Finally, according to the analysis of the Internet growth in the scale and hierarchical characteristic, a hierarchical model based on birth and death process, called BDHM shortly, is proposed. Different from the traditional models, the business relationship between AS and the exponential growth both are considered in BDHM. BDHM layers network through redundancy nodes algorithm of the greatest degree. And through computing node's birth probability by birth and death process, BDHM achieve exponential growth of network in uniform time steps. In addition, through birth and death process theory, BDHM gets the limit with time of characteristic path length. The simulation shows that the topology characteristics generated by BDHM model are all consistent with empirical evidence to some extend. And the clustering characteristic and characteristic path length both have better performance than tranditional models.