EDIC RESEARCH PROPOSAL 1 Towards Adaptive, Flexible, and Self-tuned Database Systems

Abstract—Modern database management systems (DBMSs) are used to answer complex queries on large data sets. The available physical design features along with the complexity of the modern workloads have turned the database physical design into a very complicated task. Manual design is no longer an option and the need for automating physical design tools has become more demanding than ever. DBMSs provide automated physical design in order to maximize performance and reduce the total cost of ownership. In this paper, first we study the steps that a typical automated physical designer [1] follows to propose a physical database design. We identify performance bottlenecks in the procedure namely the high number of expensive calls to the optimizer during the evaluation of different configurations and the extensive use of heuristic pruning to reduce the search space of alternative configurations. Then, we present C-PQO [6] an approach that generates a compact representation of the optimization space with a single optimization call per query. C-PQO eliminates the overhead of calling the optimizer again and again during the evaluation of different configurations and improves performance by 30x to over 450x. We also present CoPhy [13]. CoPhy introduces a combinatorial optimization formulation for the index selection problem. Thanks to the convex property of the proposed formulation, CoPhy solves the problem without heuristic pruning of the search space and can predict the quality of the final solution