Data Dependencies Preserving Shuffle in Relational Database

Abstract

This paper addresses the problem that database shuffling algorithms do not preserve data dependencies. We introduce an approach for preserving functional dependencies and data-driven associations during database shuffle. We use Boyce-Codd Normal Form (BCNF) decomposition for preserving functional dependencies. Given a relation R that is not in BCNF form, we recommend to decompose R into BCNF relations R1, ..., Rn. Each Ri (i = 1, ...,n) is shuffled then rejoined to create the shuffled relation. Our approach guarantees losslessness and preserves functional dependencies. Data-driven associations may also be lost during database shuffling. For this, we generate the transitive closure of attributes that are associated. We require that the associated attributed are shuffled together. We also present our theoretical and empirical results.