Qualitative modeling and analysis of IT infrastructures

OData support
Dr. Pataricza András
Department of Measurement and Information Systems

Nowadays, more and more IT applications leverage the cost-effective solution of the cloud environment with its rapid elasticity capabilities and on-demand scalable resources made possible by virtualization.

However, cloud infrastructures are complex systems with their own cloud-specific problems and new challenges compared to traditional dedicated solutions. Resource configurations can change during runtime, hence guaranteeing the quality of service (QoS) agreed in the service level agreement (SLA) is a much more complex task. For service critical applications, SLA violations can have serious consequences.

In order to keep QoS levels and prevent SLA violation, the system and application are monitored, and preventative corrective actions are applied if necessary. To support this, an abstract model of the system is created from measurement data and available domain knowledge. This abstract qualitative model uses qualitative instead of quantitative values, and contains the qualitatively important states and dynamics of the system while abstracting away other numerical details, which results in a smaller and easier to handle state-space.

The model can be used for decision making and for detecting critical states of the system where a preventive action (e.g. resource allocation) would avoid SLA violation and ensure the QoS levels agreed in the SLA. In addition, the same qualitative model can be used for different configurations and parametrization of the system. Furthermore, it allows the qualitative analysis and reasoning of the system, and formal checking of capacity configurations, planned policies and rules for ensuring QoS levels.

In this thesis, the abstract model was based on Qualitative Reasoning (QR), thus I examined and evaluated the Garp3 QR formalism, tool and reasoning engine, proposed an approach for building the QR model, and for using the QR model and abstraction for our tasks. Finally, I suggested extensions for Garp3 QR, and showcased parts of the proposed approach on a small set of variables from an example system.


Please sign in to download the files of this thesis.