Modeling and analysis of the internet topology

Over the last decades, the Internet has grown from a smallscale academic network linking few universities, to a largescale complex system, reaching out to billions of people and enabling communications and computing worldwide. While the number of persons connected to the Internet 24/7 is growing exponentially, particularly through smartphones, most of them are unaware of the real Internet infrastructure. Recently, researchers have attempted to discover details about the characteristics of the Internet in order to create a model of its infrastructure, which could be exploited both to improve the performances and to identify and address possible weaknesses of the network. Despite several efforts in this direction, currently no model is known to represent the Internet effectively, especially due to the lack of understanding of the true driving forces behind the Internet evolution, and the excessively coarse granularity applied by the studies done to date. This thesis seeks to scientifically understand the driving forces lying behind the exponential growth of the “network of networks”, through a fine-grained analysis of the available topological data. In the first instance, we show that the largescale Internet can be broken down into a not-so-large core, which captures the most important structural properties of the Internet, and a periphery, representing the “tendrils” of the topology. The proposed decomposition technique is general and can be extended to other networks: multi-layer network analysis, a hot topic in the area of complex networks, can benefit from it as well. We point out the key role of Internet eXchange Points within the core of the Internet, and reveal through an economic analysis the emerging competition between them and the classical Network Service Providers. This analysis serves as ground truth to devise a network model able to represent the Internet’s core effectively and potentially forecast its evolution. The model takes into account multilayer interaction mechanisms, and can be further extended to the whole Internet, by devising simple attaching mechanisms for the periphery. Since the prediction properties of the model are based on data from real measurements, it is important to remove the biases introduced by the measuring infrastructure, in order to predict the future evolution of the real Internet topology. In the end, we show through a meaningful set of metrics, how the model is able to successfully capture the Internet’s statistical and structural properties, outperforming existing topology generators in the literature.