Collaborative fault diagnosis in large-scale mobile communication networks.

摘要

With the explosive growth of mobile communication services, effective management of large-scale mobile communications networks is critical to meet customer demands and expectations. One important aspect of the network management is to provide tools that can assist fault detection and localization. Existing fault diagnosis solutions, however, most focus on addressing specific faults by leveraging ad hoc rules, and thus is quite limited in practice for managing large scale wireless networks that are dynamic and complex.;In this dissertation, we first present a system-level approach, called model-based self-diagnosis (MODI), for fault diagnosis in IEEE 802.11 wireless LANs. The MODI-embedded wireless APs can detect both security attacks and troubleshoot performance problems, and MODI can be deployed with enterprise switch-based WLAN architecture to cooperatively diagnose cross-AP problems, such as those caused by device mobility. The innovation of MODI lies in its structurally sound rule sets that identify both causality relations and expected behaviors, thus providing a holistic and practical approach for network troubleshooting.;The fault management for wide area wireless networks, such as the cellular networks, is particularly challenging given the extreme scale, complexity, dynamism and geographically distributed network components. It is cost prohibitive to deploy separate fault management infrastructure like MODI to cover the whole network. Instead, we propose a novel customer-level crowdsourcing approach that leverages the end-user mobile devices as the light-weight monitoring agents that collect user-perceived network performance metrics. We then use statistical learning techniques to uncover the relationship between the network problems with the collected performance measurements, and develop a decision tree based diagnosis model for root cause identification.;This dissertation proposes practical fault diagnosis approaches for large-scale mobile communication networks, and the experimental results show that these methods work well for both enterprise IEEE 802.11 WLANs and wide area cellular wireless networks.