An Intelligent Framework for Predicting Shifts in the Workloads of Autonomic Database Management Systems

Autonomic computing systems are intelligent systems that manage their own performance. An important characteristic of these systems is an awareness of their environment, particularly their workloads. For a complex system such as the database management system (DBMS) to be self-managing, it should be adaptive to the type of the workload put upon it. Identifying the workload type is key to tuning a DBMS and adjusting its resource allocation. We previously proposed a workload classification methodology that automatically recognizes the workload type and assesses each type's concentration. Since a DBMS may experience changes in the type of workload it handles during its normal processing cycle, it is not enough for autonomic DBMSs to identify the current type of the workload, but also to predict when a change in the workload type will occur. We could simply keep the workload classifier activated and monitor the system constantly to detect significant shifts in the type of workloads. However, this approach imposes undesirable overhead. In this paper, we propose the Psychic- Skeptic Prediction framework (PSP) that allows the DBMS to forecast major shifts in the workload by combining off-line and on-line prediction methods. Experimental results show that the PSP outperforms other possible operation modes. Furthermore, our approach is generic and can be applied to other similar prediction problems. identification methodology that automatically identifies the workload type and assesses each type's concentration using data mining classification techniques �1)�2). We build a workload classifier by analyzing a number of resource- oriented performance measures collected from the DBMS. This workload classifier can be used to identify any workload sample collected over a small time interval. The primary output of the classifier is the DSSness index, which is the percentage of the DSS type vs. the OLTP type in the workload. A DSSness of 80% means that 80% of the workload is classified as DSS and 20% as OLTP. Identifying the type of the workload, however, is just the beginning. A DBMS may experience changes in the type of workload it handles during its normal processing cycle. For example, a bank may experience an OLTP-like of short debit/credit transactions for most of the month, while in the last few days of the month, the workload becomes more DSS- like with the need to produce financial reports and run long executive queries to produce summaries. In the money market, it also has been observed that traders may exhibit some daily pattern as they access the information systems of their brokers