Spectral Analysis of Backbone Networks Against Targeted Attacks

Network science has been a central focus to correctly model and study resilience characteristics of communication networks. There have been many metrics used to represent connectivity of graphs; however, they do not suffice to compare networks with different numbers of nodes and links. The normalized Laplacian spectra enables network scientists to analyze network structures beyond what traditional graph metrics lacks. In this paper, we study the normalized Laplacian spectra of five backbone networks against targeted attacks. The physical and logical level of four commercial and one research backbone provider networks is studied. The intelligent attacks are modeled based on important graph centrality metrics of betweenness, closeness, and degree. Our results indicate that spectra of eigenvalues converge to zero after attacks. Moreover, we also identify that while in some scenarios different centrality-based attack strategies yield identical eigenvalue distribution, in other scenarios different attacks yield different eigenvalue distributions.