Evaluation and improvement of network resilience against attacks using graph spectral metrics

Measuring and improving communication network resilience against targeted attacks and random failures is an important aspect of network design. There are a plethora of proposed graph metrics to predict network resilience against such attacks. In this paper, we investigate a set of graph spectral-robustness metrics and evaluate their accuracy in predicating network resilience against node attacks. We use two datasets of graphs: baseline and random graphs to study these graph spectral robustness metrics. For each baseline graph, we apply several centrality-based attacks while we measuring the network resilience in terms of flow robustness. Using a large number of random graphs, we show the accuracy of each robustness metric to predict the graph resilience against such attacks. Furthermore, we improve the topology resilience of three real-world physical graphs via adding a set of links to maximize a given spectral metric. Among these studied metrics, we show that the network criticality robustness metric is the most accurate in predicting the behavior of a given network against centrality-based attacks. Moreover, we show that maximizing network criticality of a given graph yields the most resilient network against such attacks.