Databases Reference
In-Depth Information
tasks. This section provides an overview of the developments in mapping gener-
ation since the very first need of data transformations, until the development of
the first schema mapping tools under the form they are widely understood today.
Having defined the data exchange problem, this section describes how a mapping
scenario can be constructed. The presented algorithm, which is the basis of the
Clio [
Popa et al. 2002
] mapping scenario generation mechanism, has the additional
advantage that generates scenarios in which the mappings respect the target schema
constraints. In that sense, generating the target instance can be done by taking into
consideration only the mappings of the mapping scenario and not the target schema
constraints. This kind of mappings are more expressive that other formalisms such
as simple correspondence lines [
Rahm and Bernstein 2001
] or morphisms [
Melnik
et al.
2005
].
3.1
The First Data Translation Systems
Since the beginning of data integration, a major challenge has been the ability to
translate data from one format to another. This problem of
data translation
has
been studied for many years, in different variants and under different assumptions.
One of the first systems was EXPRESS [
Shu et al. 1977
], a system developed by
IBM. A series of similar but more advanced tools have followed EXPRESS. The
TXL language [
Abu-Hamdeh et al. 1994
], initially designed to describe syntactic
software transformations, offered a richer set of operations and soon became pop-
ular in the data management community. It was based on transformation rules that
were fired upon successful parsing of the input data. The problem became more
challenging when data had to be transformed across different data models, a situa-
tion that was typically met in wrapper construction [
Tork-Roth and Schwarz 1997
].
MDM [
Atzeni and Torlone 1997
] was a system for this kind of transformations that
was based on patterns [
Atzeni and Torlone 1995
].
Some later works [
Beeri and Milo 1999
] proposed a tree-structured data model
for describing schemas, and showed that the model was expressive enough to rep-
resent relational and XML schemas, paving the way for the later introduction of
tree-based transformations. A formal foundation for data translation was created,
alongside a declarative framework for data translation [
Abiteboul et al. 1997
]. Based
on this work, the TranScm system [
Milo and Zohar 1998
] used a library of transfor-
mation rules and pattern matching techniques to select the most applicable rules
between two schemas, in an effort to automate the whole data translation task.
Other transformation languages developed in parallel emphasized on the type check-
ing [
Cluet et al. 1998
] task or on integrity constraint satisfaction [
Davidson and
Kosky
1997
].
