Database automatic tuning tools are an essential class of database applications for database administrators (DBAs) and researchers. These selfmanagement systems involve recurring and ubiquitous tasks, such as data extraction for workload acquisition and more specific features that depend on the tuning strategy, such as the specification of tuning action types and heuristics. Given the variety of approaches and implementations, it would be desirable to evaluate existing database self-tuning strategies, particularly recent and new heuristics, in a standard testbed. In this paper, we propose a reuseoriented framework approach towards assessing and comparing automatic relational database tuning strategies. We employ our framework to instantiate three customized automated database tuning tools extended from our framework kernel, employing strategies using combinations of different tuning actions (indexes, partial indexes, and materialized views) for various RDBMSs. Finally, we evaluate the effectiveness of these tools using a known database benchmark. Our results show that the framework enabled instantiating useful self-tuning tools for these multiple RDBMSs with low effort by just extending well-defined framework hot-spots. Additionally, the instantiated tools provided significant improvements in execution cost of a query workload generated from benchmark query templates. Our framework is made available as an open-source and extensible testbed for the database research community, thus facilitating the further evaluation of database self-tuning strategies.
Foundations of automated database tuning