System models for goal-driven self-management in autonomic databases

Self-managing databases intend to reduce the total cost of ownership for a DBS by automatically adapting the DBS configuration to evolving workloads and environments. However, existing techniques strictly focus on automating one particular administration task, and therefore cause problems like overreaction and interference. To prevent these problems, the self-management logic requires knowledge about the system-wide effects of reconfiguration actions. In this paper we therefore describe an approach for creating a DBS system model, which serves as a knowledge base for DBS self-management solutions. We analyze which information is required in the system model to support the prediction of the overall DBS behavior under different configurations, workloads, and DBS states. As creating a complete quantitative description of existing DBMS in a system model is a difficult task, we propose a modeling approach which supports the evolutionary refinement of models. We also show how the system model can be evaluated to predict whether or not business goal definitions like the response time will be met.