Rethinking network management solutions models, data-mining and self-learning

Sammanfattning: Abstract på inmatningsspråket Network service providers are struggling to reduce cost while at the same time improving customer satisfaction. This thesis addresses three relevant underlying challenges to achievieng these goals: - managing an overwhelming flow of low-quality alarms - understanding the structure and quality of the delivered services - automation of service configuration All of the these add to an operator's operational costs since manual work is required in order to understand the alarm and service status as well as for configuring new services. We propose solutions based on domain-specific languages, data-mining and self-learning. We look at how domain-models can be used to capture explicit knowledge for alarms and services. In addition, we apply data-mining and self-learning techniques to understand the alarm semantics. The alarm solution is validated with a quantitative analysis based on real alarm documentation and an alarm database from a large mobile service provider. A qualitative analysis of the service management solutions is given based on prototypes and input from service providers. We present an approach to alarm interfaces by providing a formal alarm model together with a domain-specic language, BASS. This means that we can verify the consistency of an alarm interface and automatically generate artifacts such as alarm correlation rules or alarm documentation based directly on the model. From a baseline without any correlation, our alarm domain-model based on vendor documentation could automatically find the root-cause alarms in 40% of the cases. In the process of producing an alarm model from pre-existing alarm documentation for a commercial product, we found over 150 semantic warnings. We also propose a domain specic language called SALmon, which allows for efficient representation of service models, along with a computational engine for calculation of service status. Furthermore, this thesis illustrates how we can achieve automatic service conguration based on YANG, the domain-specific language standardized in the IETF. Prototypes show that the service domain-models can capture the semantics of service models, and automatically render parts of the service management solution. It is not always possible to capture expert knowledge in models. Therefore, we propose a data-mining and self-learning solution that learns alarm priorities from the decisions taken by the network administrators. The solution assigns the same severity as a human expert in 50% of the cases compared to 17% for the original alarm severity.