Information Technology Reference
In-Depth Information
The quality difference in a couple is an indicator on the relative scores of these
constructs. Obviously a coarse evaluation such as the one provided in the previ-
ous section will not allow us to define a fine-grained scale. However, it provides
a primitive but usable scale with 2 values
[0,1]
. Through a Condorcet-like voting
technique, it is possible to designate the best practice as the construct that has
been the most preferred in all the defect fixing suggestions. Then, the
1
score is
assigned to the best practice and the
score to the other constructs of the class.
Because of the limited information (constructs and quality indicators) avail-
able in the taxonomy, the production of more precise ratings should require more
investigation. Other scales are discussed in [1], in which we propose for example
to use an ordinal scale based on five grades (e.g.,
very bad
,
bad
,
neutral
,
good
,
very good
).
Improving the schema following a single quality requirements (e.g., simplicity,
expressiveness or evolvability) becomes an easy task. However in practice, quality
requirements are often combined and may lead to conflicting suggestions. When
combining criteria, two situations appear. In the first one, all the criteria come to
the same conclusion, i.e., there exists one best alternative construct that improves
all the requirements. In the other situation, there are conflicts between different
possibilities and we have to rely on trade-off techniques. For example, we can as-
sign a weight to the requirements and compute an average score if the rating are
properly defined (their scale is composed of a sucient number of values).
0
7Con lu on
The principle of taxonomy of defective constructs presented in this paper allows
us to refine the quality evaluation and improvement framework proposed in [1],
notably since it contributes to populating the equivalence classes. The taxonomy
is semi-empirical. It derives from good practices published in the litterature and
from modeling experience. The identified defects are probably representative of
the common practical defects of this last decade. It also provides designers with
guidelines to identify potential problems in database schemas and to apply
solutions according to quality criteria such as
simplicity
,
expressiveness
and
evolv-
ability
. It is important to note that this approach, based on the evaluation of
se-
mantically significant constructs
, does not oppose classical metrics approaches
counting atomic objects in the target schema. On the contrary, once defects violat-
ing definite quality criteria have been identified, they can be counted and weighted
(according to their severity) in order to produce detailed and global metrics.
Though the illustrations (taxonomy and example schemas) of this paper con-
cern the conceptual abstraction level only, the principles we have developed are
valid for all abstraction levels and all data modeling paradigms. A demand exists
for
relational schema
evaluation, inasmuch as software quality evaluation mainly
addresses software metrics at the code level (high level model evaluation still is
emerging). At this level, the quality criteria and the taxonomy are specific. For
example,
time
and
space performance
as well as
DDL portability
criteria may
