Degree Name

Master of Computer Science by Research


School of Information Technology and Computer Science - Faculty of Informatics


Performance tuning of relational database systems is always a challenging task for database administrator. Automated performance tuning has been proposed recently as a new approach to detect and to eliminate performance problems and to support the decisions of database administrators. This work considers one of the techniques used in automated performance tuning, dynamic vertical partitioning. Dynamic vertical partitioning of relational tables is one of the ways in which the physical structures of a relational database can be reorganized automatically in order to improve the performance of future database applications. The thesis presents how dynamic vertical partitioning can be used for the comprehensive analysis and optimisation of an adaptive reorganisation of database structures. In particular, we propose the algorithms to use to predict the future workload of the system, to analyse the characteristics of the workload, and to ¯nd a near optimal vertical partitioning of relational tables. Then, we discuss the implementation aspects of vertical partitioning with the materialized view and index-based techniques. Our contributions to automated performance tuning of relational database systems can be summarised as follows: 1. Propose a cost model to perform a detailed analysis of the costs of query and data manipulation processing over a given configuration of a relational database; 2. Propose a new algorithm for vertical partitioning of relational schemas in a data-base system with a given level of redundancies for a given workload; 3. Discuss the limitations of static vertical partitioning and propose dynamical vertical partitioning; 4. Discuss the characteristics of workload in order to predict the future workloads of the system; 5. Implementation aspects of vertical partition discussion: materialized view based and index based; 6. Discussion of the implementation of a vertical partition as a virtual view; 7. Conduct experiments to confirm the correctness of the cost model used by the vertical partitioning algorithm and demonstrate the expected performance gains from the partitioning

02Whole.pdf (786 kB)