Efficient distributed discovery of bidirectional order dependencies

Abstract Bidirectional order dependencies (bODs) capture relationships between lists of attributes in a relational table. They can express that, for example, sorting books by publication date ascending also sorts them age descending order. The knowledge about is useful many data management tasks, such as query optimization, cleaning, or consistency checking. Because the bODs specific dataset are usually not explicitly given, they need to be discovered. discovery all minimal (in set-based canonical form) task with exponential complexity number attributes, though, which why existing bOD algorithms cannot process datasets practically relevant size reasonable time. In this paper, we propose distributed algorithm DISTOD, whose execution time scales available hardware. DISTOD scalable, robust, and elastic approach that combines efficient pruning techniques candidates form novel, reactive, search strategy. Our evaluation on various shows outperforms both single-threaded state-of-the-art up orders magnitude; it can, particular, much larger datasets.