Cost-based prevention of violations of service level agreements in composed services using self-adaptation

Providers of composite Web services face the challenge of having to comply to SLAs, which are agreements governing the minimum performance that customers can expect from a composite service. In this work, a framework for optimizing adaptations of service compositions with regards to SLA violations has been developed. The framework, dubbed PREvent (Prediction and Prevention of SLA Violations Based on Events), uses techniques from the areas of machine learning and heuristic optimization to construct models for prediction of SLA violations at runtime, and to decide which adaptation actions may be used to improve overall performance in a composition instance. An optimizer component decides, whether applying these changes makes sense economically (i.e., whether the costs of violating the SLAs are bigger than the adaptation costs). If this is the case, the respective actions are applied in an automated way. At its core, PREvent is a self-optimizing system in the sense of autonomic computing, with the target of minimizing the total costs of adaptations and SLA violations for the service provider.