On characterizing behavior of large -scale networks

The Internet has become an important means of communication and information exchange. The fundamental ingredients needed for the successful design, engineering, and management of complex networks such as the Internet are mechanisms for characterizing their behavior. However, the development of such mechanisms is complicated by the exponential growth of the Internet and its decentralized structure. This combination of its size, rapid growth, and decentralization pose tremendous challenges to understanding the behavior of the Internet. First, since any one organization has administrative access to only a small fraction of the nodes inside the Internet, and commercial factors often prevent organizations from sharing internal performance data, the network users can not rely on network support to evaluate network-internal characteristics. Second, due to the exponential growth of the Internet, there are the needs of performance evaluation methodologies that can keep up with the rapid increasing network. The thesis exploit the use of end-to-end measurement to perform network tomography on general network topologies. We present not only several computational efficient heuristics for organizing end to end measurements but also inference techniques that obtain network-internal characteristics using end to end measurements. The thesis also develop and evaluate network analysis techniques that can efficiently explore alternative choices of network design and network engineering. It explore the use of fixed-point models to evaluate the performance of a large scale network consisting of AQM (Active Queue Management) routers that support a large population of TCP flows. Last, the thesis examine the effectiveness of existing Internet topology generators, demonstrate that none of them produces representative Internet topologies. With this in mind, the thesis propose an improved topology generator