Hora: Online Failure Prediction Framework for Component-based Software Systems Based on Kieker and Palladio.
Teerat Pitakrat · 2013
Predicting failures in large systems at runtime is a challenging task as the systems usually comprise a number of hardware and software components with complex structures and dependencies. The state-of-the-art techniques approach the task of failure prediction either by creating one separate prediction model for each crucial parameter, or by aggregating parameters of all components in order to build one prediction model. However, both approaches are not suitable for large-scale dynamic systems. In this paper, we propose our envisioned approach Hora—an online failure prediction approach for large-scale component-based systems. Our approach creates a prediction sub-model for each component and combines them using component dependencies obtained from an architectural model, e.g., PCM or SLAstic, into a model that can predict failures of internal and external services at runtime. The framework is built on the Kieker monitoring framework and other supporting tools, e.g., Weka, R, ΘPAD, which allows suitable prediction techniques to be employed for different components of the system.