Methods based on interdependent networks to analize the robustness of the internet

Abstract

The Internet allows us to communicate, access online education, commerce, etc. Its relevance has become even more apparent in recent times as we have seen the need to rely more on online communication and services. In order to ensure its proper functioning during adverse events we must study and understand the Internet's robustness. There are different ways to study this, depending on the field of study approaching the subject. Here, we use a complex networks approach. In this work we present and evaluate a newly proposed physical-logical interdependent network model inspired by today's Internet. This model considers a logical network inspired by the Internet network (Autonomous System level network), a physical network inspired by the physical Internet network (Internet backbone), and the interactions and dependencies between both networks. We propose a robustness measure to assess the model's robustness, and use it to test the effect that different types of physical attacks can have over such a system. Furthermore, we propose a novel way to attack complex networks that could allow us to better represent the effect that natural catastrophes, such as earthquakes, could have over the Internet s robustness. The main contributions of this thesis work are: (1) the development of a physical-logical interdependent network inspired by the Internet, and its characterization under different types of physical damage. (2) The finding of bridge nodes in the logical network, their effect on the overall robustness of the physical-logical interdependent networks tested, and their relation with hubs in Scale-Free networks. Our results show that finding and protecting bridge nodes can dramatically improve the robustness of a system. (3) The analysis of the effect that adding links to the physical network has over the robustness of the presented interdependent network model. (4) The development of a novel way to attack complex networks: Localized Attacks with Probabilistic Failures (LAPF). These attacks damage network elements following a probability distribution $F$, and can be used to model the damage caused by natural catastrophes. In this work we show how LAPF can be used to model the damage caused by earthquakes, and test these attacks over the physical-logical interdependent network model proposed. Our analysis shows that when studying the robustness physical-logical interdependent networks such as the one presented here, we must pay especial attention to the presence of ``bridge nodes'' as these nodes are related to events that can damage a great part of the system, even resulting in total system failure. Our results show that adding more links to the physical network can be useful to reduce the impact of bridge nodes. However, these results also show that physical link addition is not enough and adding more interlinks far apart from each other may be a better solution.