Databases Reference
In-Depth Information
There are many ways in which such a transformation can be implemented. Often,
this is done in a rather procedural fashion, and developers are forced to write quite
a lot of code to glue together the various sources. To give an example, in an ETL
application a developer would be forced to manually construct a script made of
potentially large number of simpler data-transformation steps. In other cases, such
as commercial EII systems, transformation steps are often expressed using program-
ming language (such as Java). This procedural style of specifying the mapping has
made the problem of exchanging data across different repositories quite a burden,
as discussed in
Haas
[
2007
].
To alleviate developers from this burden, we can identify two key requirements
that a mapping system should have:
A first key requirement is represented by
ease of use
and productivity. Developers
should not be required to manually specify all of the details about the map-
ping; on the contrary, users would like to specify only a high-level, abstract and
declarative representation of the mapping; then, based on this input, the mapping
system should be able to generate the actual mappings, by working out the miss-
ing details. To support this process, mapping systems usually provide a graphical
user interface using which developers may specify the mapping as a set of
value
correspondences
, i.e., correspondences among schema elements. In our example,
the input provided to the mapping system would be that shown in Fig.
5.2
;
A second essential requirement is concerned with the generation of the target
instances, i.e., with the
quality
and efficiency in the generation of solutions.
In this respect, database researchers have identified two main problems: (1) the
first one is that of
schema mapping generation
, largely inspired by the seminal Clio
papers [
Miller et al. 2000
;
Popa et al. 2002
]; this is the problem of generating a set
of mappings based on the correspondences provided as input by the user; (2) the
Fig. 5.2
An abstract specification of the mapping as a set of correspondences (
dashed arrows
denote foreign-key constraints)
