The Psychic-Skeptic Prediction framework for effective monitoring of DBMS workloads

Self-optimization is one of the defining characteristics of an autonomic computing system. For a complex system, such as the database management system (DBMS), to be self-optimizing it should recognize properties of its workload and be able to adapt to changes in these properties over time. The workload type, for example, is a key to tuning a DBMS and may vary over the system's normal processing cycle. Continually monitoring a DBMS, using a special tool called Workload Classifier, in order to detect changes in the workload type can inevitably impose a significant overhead that may degrade the overall performance of the system. Instead, the DBMS should selectively monitor the workload during some specific periods recommended by the Psychic-Skeptic Prediction (PSP) framework that we introduce in this work. The PSP framework allows the DBMS to forecast major shifts in the workload by combining off-line and on-line prediction methods. We integrate the Workload Classifier with the PSP framework in order to come up with an architecture by which the autonomous DBMS can tune itself efficiently. Our experiments show that this approach is effective and resilient as the prediction framework adapts gracefully to changes in the workload patterns.