Adaptive mechanisms for failure detection and management in complex network topologies

Abstract

Complex networks emerge naturally or as a design decision, in computational solutions. In both cases, the resulting topologies points to networks with high efficiency with respect to information dissemination and robustness, both locally and under node failures. These characteristics are credited to a small relative distance between the network elements and to a high clustering coefficient - the likelihood of neighbors connected to each other is high. Fault tolerance means that although elements fail or disconnect from the network, it can keep its main topological properties. However, studies have shown that this does not apply to cases where central elements fail continuously and concurrently. In these cases, the network may move to a completely different state, compromising its operation. Therefore, it is necessary to develop mechanisms to detect topology configurations considered critical, and promote adjustments that would reverse these states. This task is non-trivial because the topology is a result of the dynamics of the (mobile) elements and/or environmental conditions and infrastructure, which may also change over time. Moreover, the size of these networks coupled to operational constraints make methods based on global information prohibitive. Thus, this research aims to propose local mechanisms for detecting and mitigating critical topological states.