Information Technology Reference
In-Depth Information
stored [4]. The third approach (hybrid) makes use of both data sources and user
requirements [16]. With this approach, user requirements which cannot be sat-
isfied are noticed in earlier stages. Once the information from both worlds is
collected, the incompatibilities have to be solved by acommodating both data
sources and requirements in a single model.
However, with the hybrid approach a new problem arises. In the top-down and
bottom-up approaches, every element used for the implementation of the DW
comes from a single source only (either requirements or data sources), thereby
allowing us to trace elements by name matching. Nevertheless, in the hybrid
approach, additional effort is required in order to check which parts of the DW
match, not only with each requirement, but also with each part of the data
sources. Due to our experience, by following the hybrid approach, changes are
done almost in every project, since it is very common that user requirements
and data sources do not match, thus losing the implicit traceability.
In this process, the relationships between the elements are not recorded and
lost, since there is no explicit traceability included in the development process.
In turn, this hurts requirements validation [24, 26, 29], making unable to check
the current status of each requirement or take decisions about alternative imple-
mentations if a given requirement cannot be fulfilled. Although the traceability
aspect has been thoroughly studied [1, 5, 11, 22, 2, 8, 9, 23, 26], it has been almost
completely overlooked in DW development. To the best of our knowledge, the
only references to requirements traceability in DW are those from [15], which
only mention implicit traceability by name matching.
In our previous works [14, 15, 16, 17], we defined a hybrid DW development
approach in the context of the Model Driven Architecture (MDA) framework
[19]. DWs are sensitive to be developed by using MDA, cutting development
time and making the process less error prone, since transformations from the top
layer to the final implementation are performed in an semi-automatic way. In
our approach, requirements are specified in a Computation Independent Model
(CIM) by means of a UML profile [15] based on the i* framework [30]. Then,
they are automatically derived, reconciliated with the data sources in a hybrid
model, and refined through a series of layers (Platform Independent Model (PIM)
layer and Platform Specific Model (PSM) layer) until the final implementation
is achieved, as seen in figure 1.
The automatic derivation is done by means of model to model transformations
specified by Query/View/Transformation (QVT) [20] rules. QVT is a language
defined by the Object Management Group (OMG) and proposed as a standard to
create model to model transformations. However, due to our experience in real-
world projects, the lack of traceability does not allow us to adequately validate
requirements and incurs in additional costs when requirements change.
In this paper, we complement our previous works with the inclusion of the first
traceability metamodel for DWs and an automatic derivation of the correspoding
trace models. In this way, by including traceability, we improve the reusability,
maintainability and rationale comprehension of the models [24, 29], and we are
