Towards workload shift detection and prediction for autonomic databases

Due to the complexity of industry-scale database systems, the total cost of ownership for these systems is no longer dominated by hardware and software, but by administration expenses. Autonomic databases intend to reduce these costs by providing self-management features. Existing approaches towards this goal are supportive advisors for the database administrator and feedback control loops for online monitoring, analysis and re-configuration. But while advisors are too resource-consuming for continuous operation, feedback control loops suffer from overreaction, oscillation and interference. In this position paper we give a general analysis of the parameters that affect the self-management of a database. Out of these parameters, we identify that the workload has major influence on both physical design of data and DBMS configuration. Hence, we propose to employ a workload model for light-weight, continuous workload monitoring and analysis. This model can be used for the identification and prediction of significant workload shifts, which require autonomic re-configuration of the database.