Avoiding Pitfalls in Data Modeling - High-Performance Web Databases: Design, Development, and Deployment - page 291

Exhibit 18-18. Recursive relationship example of a parts explosion.

supervisor, and one supervisor supervises many employees. Mandatory
optionality, as shown in Exhibit 19b, must be expected because it implies
that everyone has to work for somebody. That rule does not reflect reality,
however, because senior-level employees do not report to anyone.

In the earlier discussion of roles, a
role entity was described as a more generalized entity. In Exhibit 8, a choice
was given between the use of a generic name entity and the use of a specifi-
cally named entity. When a given occurrence of an entity could play different
roles at different times or at the same time, however, the use of a specific
name will not work. For example, it is not possible to replace the recursive
model of the employee entity with a pair of entities like manager and subor-
dinate, because the same person could play both roles at the same time.

A Rationale for Recursive Relationship.

NORMALIZATION

Normalization of data is a procedure used to ensure that a data model
conforms to some useful standards. These standards are developed to
avoid duplication of data and to minimize create, update, and delete anom-
alies of data.

The traditional approach has been to construct the model first and arti-
ficially normalize it afterward. In contrast, proper modeling practice will
result in a naturally normalized model, at least to third normal form. The
principle to remember is that a normalized data model is of high quality,
and it is better to have quality built in than to have to add it later.

Proper data modeling practice includes the proper selection of entities,
the proper placement of attributes in entities, the recognition of when it is
necessary to make an entity of an attribute, the recognition of when to keep
information about relationships, and last but not least, the proper identifi-
cation of key attributes to ensure that all attributes depend on the whole
key and the key alone.

In all situations discussed in this article, effective data modeling prac-
tice with normalization considerations produces models that require less