beverly@ai.cs.wisc.edu (Beverly Seavey) (04/01/89)
A while ago someone claimed that normal form was not always the best form for a relational schema, and referenced an article that discussed this. Does anyone have the article reference?
edrbtsn@iuvax.cs.indiana.edu (Ed Robertson) (04/01/89)
In article <7387@spool.cs.wisc.edu> beverly@ai.cs.wisc.edu (Beverly Seavey) writes: > > A while ago someone claimed that normal form was not always > the best form for a relational schema, and referenced an > article that discussed this. Does anyone have the article reference? There are several normal forms. Which do you mean? There are two reasonable interpretations of your question: 1. The normal form in question is first normal form, requiring that all entries are atomic. There is growing interest in 'non-first normal form' databases, also called 'nested relational databases' because tuple elements can be relations rather than just atomic values. These databases provide advantages of both (standard) relational and hierarchical approaches. The original paper in this area is Makinouchi, "A consideration on normal form of not-necessarily- normalized relation in the relational data model," 1977 VLDB. There are lots of more recent works on various aspects of N1NF databases. One that sepcifically discusses design issues is Ozsoyoglu and Yuan, "A design method for nested relational databases," Proc. Third Int'l Conf on Data Engineering (1987). 2. The normal form in question is based on dependencies of some kind. A standard example on this involves the dependency {City, Street} -> ZIPCode. It is uncommon to decompose addresses with ZIPCode going into a seperate relation (although doing so might, for example, ease the transition from a 5-digit to a 9-digit code). Maybe htis comes from Date's book (someone has borrowed my copy so I can't check). -- Edward Robertson robertson@cs.indiana.edu Computer Science Dept Indiana University 812-855-4954 Bloomington, IN 47405-4101