End-to-end service survivability under attacks on networks

Network survivability is a capability of a networked system to provide its services despite failures or attacks. Attacks, e.g., due to acts of war, being potentially damaging events, were basically considered in the historical definitions of a survivability phenomenon. The meaning of the term: "network survivability" evolved in the last decade. Recently, attacks replayed the important role again. Their nature, however, including intrusions, probes, denials of service, differs from the old one. Survivability is strongly related to other fields of study. In particular, quality of service depends on network survivability. We investigate these dependencies in scale-free networks. Many networks are scale-free, i.e., their node degree distribution follows the power law. Nodes of the highest degrees, called centers, are highly vulnerable to attacks. Elimination of these nodes seriously degrades the overall performance of network services. In this paper we propose a model, which, based on traffic parameters of a demand, like delay or bit rate, allows to establish the survivable and attack proof end-to-end connections. The key idea of this model is that for the significant traffic, it establishes paths, which omit centers. The important connections become more resistant to attacks. We show that in the best case, obtained for the highest class of service, the number of broken connections is reduced even by factor 3. Example results are compared to those for the standard distance metrics. Our model is applicable to many network architectures and many classes of service.