Information Technology Reference
In-Depth Information
types) may be, in some circumstances, considered a defect since a mere many-
to-many relationship type would better express the intention of the designer.
Considering a set of about 20 basic conceptual patterns, we have defined as
many equivalence classes, each of them gathering all the patterns that express
the same semantics. For example, the relationship entity type and many-to-many
relationship type patterns appear in the same equivalence class. In each class,
we can identify its representative member, that is, the pattern that best ex-
presses the common intention of the members of this class (and for this, called
best practice
). For example, the semantic pattern
many-to-many association
will
be described by a class that includes, among a dozen equivalent patterns, the
many-to-many relationship type, the relationship entity type, the multi-valued
foreign key, the multi-valued embedded component. Clearly, the first pattern
will be the representative member of this class. We will see in the section 3, that
the best practice of an equivalence class depends on the quality criteria for the
evaluation of which this class is used.
The qualification
defective
of a construct is not absolute
3
but depends on three
factors, namely the abstraction level, the modeling paradigm and the quality
criterion. For example, at the logical level, the
foreign key
, as the expression
of a many-to-one relationship type, is optimal in a class of logical constructs
but sub-optimal in a class of conceptual constructs. It is optimal in the SQL
paradigm but not in the ADO Microsoft interface, based on a simple Entity-
relationship model. It may be considered sub-optimal in an XML schema where
element embedding may be preferred for performance reason.
In this paper, we will deepen the framework by exploring the space of concep-
tual defects and by attempting to classify them into an ontology of natural defect
types. These reference defect types contribute to a better understanding of the
third factor mentioned above: quality criterion. This classification will be used to
improve our quality evaluation framework, but it has also been used in database
design education [2] in the perspective of building high quality schemas.
Since most, if not all, database schemas include a certain amount of defects
and considering that database design mainly is a creative task, we can expect the
catalog of schema defect types being very large. In the following sections, we will
concentrate on defects that degrade otherwise correct schemas. For example, a
relational table that is not in 3NF is not intrinsically incorrect but it leads, among
others, to expressiveness (two fact types are represented in the same table) and
performance (space and update time) problems. The process of identifying these
defects and improving their
structural
quality is generally known as
Conceptual
normalization
.
The paper will be structured as follows. Section 2 presents a short state of the
art in the role of defects in database schema quality. We recall the main concepts
of the framework in section 3. Section 4 describes the bases of quality analysis
for conceptual schemas. In section 5, we present a taxonomy of conceptual data
schema defects and discuss their improvement. The use of the framework extended
by this taxonomy is presented in section 6. Section 7 concludes the paper.
3
For this reason, we have avoided the term
anti-pattern
.
