Nobody can deny the importance of an Internet that has grown dramatically over the past few years. Nowadays, data has become a strategic asset for individuals and organizations alike; this means that even a partial failure of a small part of the Internet is no longer just an annoyance, but can have serious consequences for many people. This paper addresses the effects of such failures and then proposes a way to mitigate them.
Of course, the paper cannot take into account all the devices connected to the Internet, so it concentrates on AS-level nodes, or sets of Internet protocol (IP) addresses owned by large single administrative entities such as Internet service providers (ISPs). Each of these nodes has a degree, or how many other nodes it is linked to; nodes can be grouped in networks with a k-core index, which denotes the minimum degree among all nodes in the network. The main goal of the paper is to measure network robustness, or how well the network continues transmitting data while nodes go offline because of failures, either intentional or not, and are brought back online thanks to operator efforts. Up to this point, the paper tackles this topic in theory; now, it turns its attention to real-world data sets taken from the DIMES project (http://www.netdimes.org) and the UCLA AS-level topology archive (http://irl.cs.ucla.edu/topology). It simulates attacks on these data sets and finds that the higher the k-core index of a network, the less damage is caused by failures, because many paths are in any case left open; however, even three-to-six percent of nodes out of order are enough to crash the network. To counter this effect, the final part of the paper proposes ELL, a methodology to enhance the robustness of a network by adding more links between nodes of low degree.
I expose here the content of the paper in a narrative way for the sake of simplicity, but of course it presents the same topics, from k-core calculations to attack simulations to discussing results, in statistical and mathematical terms.