Grant: Collaborative Research and Development Grants

The current trend for enterprises to consolidate different workloads, such as business intelligence (BI) and transaction processing (OLTP), with different performance requirements onto a single database server means that the management of the database management system (DBMS) is becoming increasingly complex and costly. The competition among the workloads for server resources, for example, must be managed based on the importance and the demands of the workloads. BI workloads are especially resource-intensive and are often of lower importance than OLTP workloads. We must therefore be able to ensure that BI queries make progress while the requirements of more important workloads are met. Existing workload management techniques, such as assigning operating system priorities to workloads, or terminating then restarting BI queries in order to free up their resources, are inadequate. We propose to create a comprehensive workload management framework that integrates several complementary mechanisms for workload management as well as autonomic controllers to dynamically adjust the application of these mechanisms depending upon the current performance levels of the DBMS. We will develop novel methods to govern the multiple controllers in a coordinated fashion. We will create an experimental test bed and evaluate the framework for two key scenarios, namely to avoid system overload and to meet specific workload service level objectives. The proposed research addresses key challenges facing IBM Canada and other DBMS vendors as they seek to incorporate autonomic computing features in their products. The project will produce technology that can be incorporated into future versions of DBMS products such as IBM's DB2 and Sybase's Advantage Database Server. The project will also provide training for eight HQP (a Ph.D. student, three M.Sc. students, three undergraduate students and a Research Associate).