Databases Reference
In-Depth Information
Although the normalization process is well defined and the correspond-
ing algorithms easy to implement, few existing CASE tools provide normali-
zation support for relational schemas. The common reason always given
is that normalized schemas are not optimized schemas, and the necessary
denormalization step that follows makes the process useless and time con-
suming. Obviously this argument is not acceptable because it is clear that
logical design has nothing to do with physical design. The purpose of logical
design is to build a sound relational schema that will serve as a reference for
the DB evolution, while the physical design intends to provide an implemen-
tation that optimizes the set of actual queries. Physical design is subject to
change more frequently than logical design, as the application queries and
technology evolve.
The main problem of the normalization theory is the acquisition of
dependencies. Functional dependencies are semantic assertions stated about
attributes. Although inference rules between dependencies [18] allow us
to derive some dependencies from others, elementary dependencies cannot
be defined without the users support. This is probably the most important
obstacle to the use of normalization tools. The next section investigates dif-
ferent techniques that can support functional dependencies acquisition and
can be added to the normalization process to make it more pragmatic.
13.4.2
Functional Dependency Acquisition
Given that the number of possible functional dependencies between n attrib-
utes is the function of n!, the acquisition of these dependencies is the bottle-
neck of the normalization process. A question-answering system becomes
useless without intelligent techniques to dramatically reduce the number of
questions for the users or designers. To deal with the combinatory aspect
of this problem, we present a panel of heuristics and techniques that reduce
the search space of functional dependencies.
Removing the Universal Relation Assumption
The universal relation assumption allows the definition of functional
dependencies in the absence of existing relations. In the case where the rela-
tional schema is derived from the conceptual schema by mapping rules, as we
will see in Section 13.4.3, there is no need to have this assumption. Indeed,
functional dependencies can be defined on first normal form (1NF) relations
that resulted from the transformation of the conceptual schema. Each of
these 1NF relations can be considered a universal relation over which func-
tional dependencies must be defined. The normalization algorithm will
