Autonomic index selection in relational database management systems by evolutionary transformation of query execution plans

Aim of project
Relational database management systems (RDBMS) have been under continuous development for over three decades. Their complexity makes the administration process requires much experience and highly specialized knowledge. Any methods for reducing the costs of administration are nowadays in great value. These methods usually employ some fully or partially autonomic components. One of the common tasks required for autonomic operation of the RDBMS is optimal index selection. The index selection process must not impose any significant runtime and memory overhead on the operation of the RDBMS. The state-of-the-art greedy index selection heuristics often do not provide solutions of sufficient quality, while the global optimization methods leveraging the “what-if” approach are not efficient enough in autonomic applications. The aim of the research is to create a method that provides better and faster autonomic index selection than the state-of-the-art algorithms. The method should be applicable to database workloads that change their characteristics over time. The preliminary experiments have shown that the evolutionary searching of the space of query execution plans (as opposed to commonly used searching of the space of index configurations) yields fast convergence to good solutions. The proposed algorithms will be theoretically analyzed, implemented in Java and experimentally evaluated for various real-world and synthetic database workloads. The results will be compared to the methods currently used in commercial RDBMS

Expected results
Expected results: The proposed method should reflect a better efficiency than other methods. It is also expected that the results will contribute not only to developing a new method for the selection of indexes, but will also become the basis for the development of new algorithms selecting other physical data structures in the RDBMS in the future.