Robustness analysis to multiple failures in GMPLS networks

Natural disasters such as earthquakes and flooding, as well as massive blackouts and other large-scale accidents can all seriously disrupt the normal operation of communication networks. Such events potentially affect several network elements simultaneously, thus creating so-called multiple-failure events. This paper studies the robustness of GMPLS-based networks subjected to such multiple-failure events from the perspective of the survival of end-to-end connections when subsets of links are made invulnerable against the specific failure, i.e., when immunization is applied. Two immunization strategies for improving robustness are compared. The first one uses purely static topological data to prioritize links based on their estimation of centrality, while the second one relies on link usage information to derive a criticality measure. These strategies are evaluated through simulation on two synthetic topologies representing continental-size transport networks, on which the normal operation and the occurrence of large-scale failures are simulated. The results also show that the two immunization strategies are effective in minimizing the number of affected connections and at the same time highlight the limitation of purely topological-based measures of network robustness.