Schema optimisation instead of (local) normalisation (Q2206788)

The paper provides a useful survey of the development of normalization techniques for the design of relational database schemas. The paper's content is divided into four sections. Section 1 presents important concepts and background concerning normalization. It discusses the classical local vertical normalization based on functional dependencies, decomposition and inclusion constraints. Also horizontal decomposition and associated exclusion constraints are mentioned. The author shortly specifies six targets of normalization. The rest of the paper shows that the six targets are not achieved until now, both in theory and practice. Section 2 lists several reasons why local normalization may be inadequate. It presents some solutions for classical normalization approaches, especially the refinement of synthesis algorithms. A detailed discussion is devoted to the decision between using minimal covers and normal forms or a normalization approach for entity-relationship schemata. Section 3 is devoted to the concept of denormalization not often mentioned in the literature. Two basic approaches are discussed: (1) no denormalization at all; (2) liberal and controlled denormalization whenever it is really necessary. The basic approaches then include denormalization as the inverse of normalization and denormalization as the extension of a schema. The basic contribution of the author is proposed in the short subsection describing denormalization driven by optimization. A high-level formal apparatus as presented here should contribute to the optimization approach and consequently to meet demands for the six targets why we should normalize. It is a pity that at least a short case study is not given in this section. In Section 4, a summarization of the paper is presented, with the conclusion that normalization should be revised for modern DBMS technology. Yes, a movement to sets, multisets, lists, etc. is necessary for basic database types. For non-first-normal-form occurring now in SQL database types this is also the case. And also in the context of big data, the author calls for global normalization, i.e., a theory for system structures in the large. For the entire collection see [Zbl 1435.68031].