Fault diagnosis in network virtualization environment

Abstract

Virtual networks have emerged as a powerful and flexible platform for future network. The dependability of virtual services relies on the network's capabilities to effectively diagnose and recover faults. But the flexible characteristics of virtual networks bring to virtualization fault diagnosis new challenges, such as network scalability, inaccessible substrate network fault information, incomplete and inaccurate network observations, dynamic symptom-fault causality relationships, and multi-layer complexity. To tackle with these challenges, the paper proposes a virtual network fault diagnosis framework (called VNFD). VNFD can use observed end-to-end symptoms reported by monitoring systems to get a set of potential faulty components for evaluating their fault likelihood, and select the most likely faulty hypothesis set to explain all the observed symptoms. VNFD can locate root causes without relying on substrate network fault probabilistic quantification (e.g. prior fault probability). Simulations and experimental studies show that VNFD can efficiently and accurately get a hypothesis set explaining all the observed symptoms, even when the symptoms are incomplete and inaccurate.