Information Technology Reference
In-Depth Information
Quality Evaluation and Improvement Framework
for Database Schemas - Using Defect Taxonomies
Jonathan Lemaitre and Jean-Luc Hainaut
Laboratory of Database Application Engineering - PReCISE research Center
Faculty of Computer Science, University of Namur
Rue Grandgagnage 21 - B-5000 Namur, Belgium
{jle,jlh}@info.fundp.ac.be
http://www.fundp.ac.be/precise
Abstract. Just like any software artefact, database schemas can (or
should) be evaluated against quality criteria such as understandability,
expressiveness, maintainability and evolvability. Most quality evaluation
approaches rely on global metrics counting simple pattern instances in
schemas. Recently, we have developed a new approach based on the iden-
tification of semantic classes of definite patterns. The members of a class
are proved to be semantically equivalent (through the use of semantics
preserving transformations) but are assigned different quality scores ac-
cording to each criteria. In this paper, we explore in more detail the
concept of bad pattern by proposing an intuitive taxonomy of defective
patterns together with, for each of them, a better alternative. We iden-
tify four main classes of defects, namely
complex constructs
,
redundant
constructs
,
foreign constructs
and
irregular constructs
. For each of them,
we develop some representative examples and we discuss ways of im-
provement against three quality criteria:
simplicity
,
expressiveness
and
evolvability
. This taxonomy makes it possible to apply the framework to
quality assessment and improvement in a simple and intuitive way.
Keywords: Conceptual data schema, quality, schema improvement,
schema evaluation, schema transformation.
1
Introduction
Modern engineering approaches to system development lead to methods in which
modeling activities have become prominent, notably through the so-called
model
driven engineering (MDE)
initiative. According to these methods, the design of
a complex software system appears as a hierarchy of models, starting from the
goal model down to the source code of the concrete artifacts. Models derive
from each other through transformations that preserve some of their intrinsic
properties, such as correctness, information capacity or performance. In addition,
most models use components of other models. Through these derivation and
use dependencies, a defect in a source model potentially propagates to many
dependent models. In such an interconnected model network, the quality of the
whole system critically depends on the quality of each of its models.
