|dc.description.abstract||This research investigates the problem of the system-level fault diagnosis in mobile wireless networks using the comparison approach. Mobile wireless networks deliver crucial services in harsh environments. Remarkably, there is a proliferate reliance upon services running on such systems. However, efficient service delivery is a substantial challenge due to the intrinsic characteristics of mobile networks and the rough deployment conditions. Hence, researchers have paid much attention to design dependable mobile networks withstanding failures that may lead to service outages. Faults are the sources of network impairments, and hence fault diagnosis is a leading mean to attain network dependability. System-level fault diagnosis has been studied widely, aiming to automate the diagnosis process. One of the most practical diagnosis approaches is the comparison approach that identifies the faulty status of nodes by comparing their return outputs for the same task assigned earlier. In the literature, there are several comparison-based diagnosis models. However, the characterisations of the diagnosable systems under these models impose stringent constraints on the underlying system, and hence their competency in dynamic contexts deteriorates sharply.
This thesis presents three significant contributions to tackle the dearth of diagnosis models proposed for mobile networks. First, it scrutinises the diagnosis requirements of mobile wireless networks. Further, it addresses the fundamental limitations of the current comparison-based diagnosis models and protocols. Second, this thesis presents a time-free comparison model for mobile wireless networks. The class of diagnosable systems under this model has been characterized. Two fault diagnosis protocols have been presented, and their performance has been evaluated. Both protocols can correctly diagnose faulty nodes undergoing static and dynamic faults in mobile wireless networks. These protocols employ two different dissemination approaches to exchange local views among nodes. The first protocol employs a flooding-based technique, whereas the second one leverages a random linear network coding technique. Third, this thesis presents a probabilistic comparison model for mobile networks. This model supports more realistic fault model where nodes can be faulty with probability. In this model, not only permanent faults are allowable in a system under diagnosing, but also intermittent faults. An efficient diagnosis protocol that implements this model has been proposed. This protocol can identify permanent and intermittent faults with high probability. The correctness proofs, analytical analysis, and performance evaluation using simulations have been presented. Detailed discussions and comparisons among models and protocols proposed to date reveal significant potentials of our proposed diagnosis models and protocols. Undoubtedly, these models pave the way for developing efficient fault diagnosis protocols for mobile wireless networks, and hence attaining the dependability with low excess overheads||en_NZ