Dependencies between Models in the Model-driven
Design of Distributed Applications
1
João Paulo A. Almeida, Luís Ferreira Pires, Marten van Sinderen
Centre for Telematics and Information Technology, University of Twente
PO Box 217, 7500 AE Enschede, The Netherlands
Abstract.. In our previous wo
rk, we have defined a model-driven design
approach based on the organization of models of a distributed application
according to different levels of platform-independence. In our approach, the
design process is structured into a preparation and an execution phase. In the
preparation phase, (abstract) platforms and transformation specifications are
defined. These results are used by a designer in the execution phase to develop
a specific application. In this paper, we analyse the dependencies between the
various types of models used in our design approach, including platform-
independent and platform-specific models of the application, abstract platforms,
transformation specifications and transformation parameter values. We consider
models as modules and employ a technique to visualize modularity which uses
Design Structure Matrices (DSMs). This analysis leads to requirements for the
various types of models and directives for the design process which reduce
undesirable dependencies between models.
1 Introduction
In our previous work [1, 2], we have defined a model-driven design approach (aligned
with the Model-Driven-Architecture [7]) based on the organization of models of a
distributed application according to different levels of platform-independence. In this
approach, models at a particular level of platform-independence can be realized with a
number of platforms (such as, e.g., middleware platforms), possibly through
application of successive (automated) transformations that lead ultimately to
platform-specific models, i.e., models at the lowest level of platform-independence
with respect to a particular definition of platform.
An important architectural concept
of our approach is that of an abstract platform.
An abstract platform is an abstraction of infrastructure characteristics assumed for
models of an application at a certain level of platform-independence. An abstract
platform is represented through metamodels, profiles and reusable design artefacts
[1]. For example, if a platform-independent design contains application parts that
interact through operation invocations (e.g., in UML [8]), then operation invocation is
a characteristic of the abstract platform. Capabilities of a concrete platform are used
1
This work is part of the Freeband A-MUSE project. Freeband (http://www.freeband.nl) is
sponsored by the Dutch government under contract BSIK 03025.
Paulo A. Almeida J., Ferreira Pires L. and van Sinderen M. (2005).
Dependencies between Models in the Model-driven Design of Distributed Applications.
In Proceedings of the Joint Workshop on Web Services and Model-Driven Enterprise Information Systems, pages 117-131
DOI: 10.5220/0002571601170131
Copyright
c
 SciTePress

during platform-specific realization to support this characteristic of the abstract
platform. For example, if CORBA is selected as a target platform, this characteristic
can be mapped onto CORBA operation invocations.
An indispensable activity in early stages of our development approach is to
determine the levels of models, the abstract platforms, and the (automated)
transformations that are needed. This activity is part of the preparation phase of the
MDA development process [6]. In the preparation phase, (MDA) experts define the
metamodels, profiles and transformations that are to be used in the execution phase by
application developers. In the execution phase, a specific application is developed
using the generalized designs and design knowledge captured during the preparation
phase.
Figure 1 shows the various models manipulated in our approach. Three levels of
platform-independence are depicted, and the results are classified according to the
phase in which they are produced. In this figure, an arrow indicates that a model is
dependent on the existence of another model by construction. Abstract platforms have
been depicted as models, indicating that abstract platform definitions can be captured
in abstract platform models. Transformation specifications have also been depicted as
models, indicating that generalized design operations can be captured and reused.
Transformation specifications can be parameterized and values for transformation
parameters are defined in the execution phase. These values are called transformation
arguments. Arguments of a transformation are also called markings when these are
associated to elements in a source model, in which case transformation parameters are
called marks.
Ideally, models in our approach (presented in Figure 1) should be independent of
each other, i.e., it should be possible to create models independently, and a
modification in one model should not impact other models. Nevertheless, models
capture design decisions on the same object of design, i.e., the same application, and
hence not all models are independent of each other. The benefits of separation of
models are reduced when models are related in such a way that modifications in a
model affect other models. In this paper, we analyse the dependencies between the
various types of models used in our design approach and strive to find techniques to
avoid undesirable dependencies between models.
application
PI M M
1
application
PI M M
2
transformation
specification T
1
transformation
ar gu ment s a
1
abst r ac t
platform Π
1
abst rac t
platform Π
2
level 1
level 2
application
PSM M
3
transformation
specification T
2
transformation
ar gu ment s a
2
concrete
platform Π
3
level 3
preparation phase execution phase
Fig. 1. Models in our design approach
118

Dependencies between models restrict the opportunities for division of labour and
concurrent design. Interdependencies reduce the efficiency of the design process and
often have to be addressed in the design process by introducing iteration cycles [4].
As we elaborate in this paper, some interdependencies can be avoided by following a
number of rules with respect to the content of the various models and with respect to
the modifications that may be applied to the various models.
In the remainder of this paper, we address the following questions with respect to
the separation of models in our approach (among others):
– can concrete platforms be modified without affecting PIMs and abstract
platforms?
– can transformation specifications be modified without affecting PIMs and abstract
platforms?
– does a modification in a PIM affects a corresponding PSM?
– does a modification in a PSM affects a corresponding PIM?
– are there interdependencies between the various models that require iterations in
the design process? Can these be avoided?
This paper is further organised as follows: section 2 proposes that models should
be considered as modules whose modularity can be analysed through a technique
called Design Structure Matrices (DSMs) [9, 10]; section 3 analyses the
(inter)dependencies between the various types of models, which results in
requirements and guidelines for the separation of models; section 4 discusses how the
dependencies between models affect the design process; section 5 classifies the
different models according to their various dependencies; finally, section 6 presents
some concluding remarks.
2 Models as modules
In order to examine the relations between the various models, we consider models as
modules. Typically, a module is a set of elements of a design that are grouped
together according to an architecture or plan, with three main purposes [3, 4]: to make
complexity manageable; to enable parallel work; and to accommodate future
uncertainty.
While modularization is often used as a technique to split up and assign different
functions of a complex system to different system parts, we split up and assign
different design decisions to different models. A number of basic principles of
modularity apply both to the functional decomposition of system parts (within a
model) and to the separation of models in our design approach.
As is noted in [4]: “a complex engineering system is modular-in-design if (an only
if) the process of designing it can be split up and distributed across different separate
modules that are coordinated by design rules, not by ongoing consultations amongst
the designers.” This definition reveals two important features of systems that are
modular-in-design:
– Independence: The absence of ongoing consultations amongst the designers of
different modules reveals that modules should be largely independent of each
other. Modules correspond to independent activities in the design process; and
119

– Dependence: The relations between the different modules are defined by a set of
design rules
2
to be respected. These design rules reflect the need for coordination
of design choices. Separating strongly related modules forces the number of
design rules to increase, constraining the freedom of designers of the different
modules.
In the following sections, we examine independence and dependence of models in our
design approach. We employ a technique to visualize modularity-in-design which
uses Design Structure Matrices (DSMs) [9, 10]. DSMs have been used extensively in
the field of Engineering Design, both for products and production processes and
design processes [4]. In this technique, modules are arrayed along the rows and
columns of a square matrix. The matrix is filled in by determining, for each module,
which other modules affect it and which are affected by it. The result is a map of the
dependencies between the modules.
3 Dependencies between models: two levels of models
We start our analysis by assuming two levels of design within a single design iteration
cycle as depicted within the rounded rectangle in Figure 2.
design activities
design activities
level 1
level 2
user requirements
design 1
design 2
design activities
design activities
design 1’
design 2’
user requirements’
design activities
design activities
design 1’’
design 2’’
user requirements’’
...
Fig. 2. Two levels of models related by transformation
We assume further that the preparation phase results in an abstract platform Π
1
for
designs at level 1, a concrete platform Π
2
for designs at level 2. The design activities
are constrained by a transformation specification T
1
that relates models that rely on Π
1
to models that rely on Π
2
. This situation is depicted in Figure 3. This figure reveals
the various models of the execution phase that are considered at this point of our
analysis, namely, an application PIM, transformation arguments, and an application
PSM.
2
In functional decomposition, interfaces between components are considered design rules.
120

app l i c at i o n
PI M M
1
appl i c ati on
PSM M
2
transformation
specification T
1
transformation
ar gument s a
1
abst ract
platform Π
1
concrete
platform Π
2
T
1
design activities
…
parametrization
…
preparati on phase
transfer of results
dependency by construction
Fig. 3. Two levels of models related by transformation
We discuss the dependencies between each of the models depicted in Figure 3 in the
following sections. In each section, we discuss how the various models are affected as
a result of a modification of one of the other models. After the relations between all
models are examined, a DSM is built to visualize the dependencies between the
various models.
Application PIM. Table 1 shows the dependencies between the various models and
an application PIM. The ‘
7’ symbol marks the existence of some dependency. The
absence of the symbol indicates there is no dependency. We justify the existence or
absence of a dependency for each pair of models.
121

Table 1. Dependencies between the various models and an application PIM
Application PIM Explanation
Application
PIM
N/A trivial
Abstract
platform
An abstract platform is designed so that it can be used to design a class of applications;
the modified application PIM is still a member of this class of applications.
This constitutes a generality requirement for abstract platform, but also sets the
constraints on possible modifications of an application PIM for a given abstract
platform.
Application
PSM
7 through
transformation
The relations between application PIMs and PSMs are determined by transformation
specifications and transformation arguments; if the application PIM is modified, it is
possible that the modified PIM and the original PSM no longer respect this relation; in
this case, the PSM or transformation arguments may be affected by change.
Concrete
platform
The concrete platform is a member of the set of target platforms implied by portability
requirements; all application PIMs that rely on the abstract platform must be buildable
(see explanation below about buildability) in the concrete platform, thus requiring no
modifications in the concrete platform.
This constitutes requirements for the abstract platform and transformation
specification.
Transf.
arguments
7
Transformation arguments are used to introduce variation in transformation
specifications, in order to capture particular design decisions; these decisions may be
application-specific or may refer to elements of the application PIM; e.g.,
transformation parameters can be used to specify the physical allocation of each
application component in the application PIM.
Transf.
specification
Transformation specifications are designed so that they can be applied to the class of
applications that can be built on top of an abstract platform; the modified PIM is still a
member of this class of applications.
This constitutes a generality requirement for transformation specification.
Buildability of a design is inversely proportional to the amount of time, effort and
resources required to build a conformant realization of the design on a particular
platform. Buildability depends on the contents of a design. The actual contents of a
platform-independent design depend partly on the abstract platform, which is defined
in the preparation phase. Therefore, in the preparation phase, buildability can only be
estimated indirectly, by analysing the impact of abstract platform characteristics in the
buildability of the class of application designs supported by the abstract platform. We
propose this is done by examining the differences and similarities in the abstract
platform and target platforms
3
.
Having introduced the notion of buildability, we are able to formulate a definition
of platform-independence of a design. We say that a design is platform-independent
if, and only if, it is buildable on a number of target platforms. The set of target
platforms is determined by portability requirements for the design, which are
themselves determined by technical, business and strategic arguments.
Abstract platform. Table 2 shows the dependencies between the various models
and an abstract platform.
3
We have explored this idea initially in [2].
122

Table 2. Dependencies between the various models and an abstract platform
Abstract
platform
Explanation
Application
PIM
7
By definition: “an abstract platform is an abstraction of infrastructure characteristics assumed
in the construction of PIMs of an application”; if these characteristics change, the application
PIM may be affected.
Abstract
platform
N/A trivial
Application
PSM
Modifying an abstract platform may affect PIMs, transformation specifications (see respective
cells in this table), which in turn may affect application PSMs (see other tables); however,
only direct dependencies are represented in a DSM.
Concrete
platform
The set of target platforms is determined by portability requirements; during abstract platform
definition, buildability with respect to the target platform must be observed.
This constitutes a requirement for abstract platform definition.
Transf.
arguments
Transformation arguments depend on the transformation specification, which depends on
abstract platforms (see cell below); however, only direct dependencies are represented.
Transf.
specification
7
The abstract platform defines the common characteristics of a class of platform-independent
designs for which there should be generalized implementation relations to different platforms;
these implementation relations are captured in transformation specifications; a change in
abstract platform characteristics changes the class of applications, invalidating assumptions on
common concepts, patterns and structures that were made to define transformations.
The separation between an abstract platform and a transformation specification is
analogous to the separation between an interface definition and a realization of the
interface in component-based design: an abstract platform defines requirements which
are satisfied by one or several transformation specifications.
Application PSM. Table 3 shows the dependencies between the various models and
an application PSM.
Table 3. Dependencies between the various models and an application PSM
Application PSM Explanation
Application
PIM
7 through
transformation
The relations between application PIMs and application PSMs are determined by
transformation specifications and transformation arguments; if the application PSM is
modified, it is possible that the modified PSM and the original PIM no longer respect
this relation; in this case, the PIM or transformation arguments may be affected by
change. This dependency exists for both unidirectional and bidirectional [5]
transformations. In the case of bidirectional transformations, changes to PIM may be
propagated automatically.
Abstract
platform
A modification in an application PSM may result in a modification in the application
PIM (see cell application PIM above); the modified PIM is still a member of this class
of applications for which the abstract platform is defined.
This constitutes a generality requirement for abstract platform, but also sets the
constraints on modifications of an application PSM for a given abstract platform.
Application
PSM
N/A trivial
Concrete
platform
A concrete platform is designed so that is can be used to design a class of applications;
the modified PSM is still a member of this class of applications.
This constitutes a generality requirement for concrete platforms.
Transf.
arguments
7 through
transformation
(see cell application PIM above)
Transf.
specification
Transformation specifications define generalized implementation relations;
transformation specifications define a class of PSMs that conform with PIMs; the
modified PSM is still a member of this class of applications.
This constitutes a generality requirement for transformation specifications, but also
sets the constraints on possible modifications of an application PSM for a given
transformation specification and a PIM.
123

Concrete platform. Table 4 shows the dependencies between the various models and
a concrete platform.
Table 4. Dependencies between the various models and a concrete platform
Concrete
platform
Explanation
Application
PIM
independence
is engineered
Independence is engineered in the definition of abstract platforms.
This constitutes a buildability requirement for abstract platforms.
Abstract
platform
independence
is engineered
Independence is engineered in the definition of abstract platforms.
This constitutes a buildability requirement for abstract platforms.
Application
PSM
7
Application PSM depends on sets of concepts, patterns and structures provided by a
concrete platform; the instability of concrete platforms, and hence application PSMs,
motivates separation of platform-independent and platform-specific concerns in our
approach.
Concrete
platform
N/A trivial
Transf.
arguments
7
Transformation arguments may be platform-specific, e.g., markings may define that
particular components should be transformed into Session or Message-Driven
Enterprise Java Beans.
Transf.
specification
7
Transformation specifications define generalized implementation relations for a
particular target platform; change the target platform and these relations may be
invalidated. Ideally, this dependency could be reduced by using concrete platform
models as transformation arguments. However, this solution requires highly general
transformation specifications, which define generalized implementation relations for a
class of target platforms (resulting in a platform-independent transformation
specification).
Transformation arguments. Table 5 shows the dependencies between the various
models and transformation arguments.
Table 5. Dependencies between the various models and transformation arguments
Transf.
arguments
Explanation
Application
PIM
Abstract platforms are defined to preserve freedom of implementation, so that different
implementations of application PIMs built on top of it are possible; since transformation
arguments are used to introduce variations in generalized implementation relations,
changes in transformation arguments should not affect application PIMs nor abstract
platforms.
This constitutes a requirement for abstract platforms and transformations, and sets the
constraints on possible modifications of transformation arguments for a given
combination of abstract platform and transformation specification.
Abstract
platform
(see cell application PIM above)
Application
PSM
7 through
transformation
The relations between PIMs, transformation arguments and PSMs are determined by
transformation specifications; if transformation arguments are modified, it is possible
that the original PIM, the modified arguments and the original PSM no longer respect
this relation; in this case, the PSM may be affected by change in transformation
arguments.
Concrete
platform
A concrete platform is designed so that is can support a class of applications; a PSM
that is affected by a change in transformation arguments is still a member of this class of
supported applications, therefore, requiring no modification of the concrete platform.
This constitutes a requirement for transformation specification, namely that the results
of transformations are always PSMs that use the concrete platform.
Transf.
arguments
N/A trivial
Transf.
specification
Transformation specifications have transformation parameters, which are assigned
values when the transformation specification is instantiated.
124

From the perspective of model transformation, the distinction between PIMs and
transformation arguments is unnecessary: both PIMs and transformation arguments
may be considered as input information for an unparameterized transformation.
However, the distinction is relevant from the perspective of the design process: PIMs
are platform- and transformation independent, while transformation arguments may
be platform- and transformation specific. Transformation arguments may be defined
after PIMs have been conceived. As a consequence, designers of PIMs may not be
aware of whatever transformation parameters may be chosen by a designer using the
PIM as a starting point to derive a PSM.
Transformation specification. Finally, Table 6 shows the dependencies between the
various models and a transformation specification.
Table 6. Dependencies between the various models and a transformation specification
Transf.
specification
Explanation
Application
PIM
Abstract platforms are defined to preserve freedom of implementation, so that different
implementations of application PIMs built on top of it are possible; these different
implementations are captured in transformation specifications.
This constitutes a requirement for abstract platform, but also sets the constraints on
possible modifications of transformation specifications for a given abstract platform.
Abstract
platform
(see cell application PIM above)
Application
PSM
7
The relation between application PIM and application PSM is determined by
transformation specifications and transformation arguments; since a change in
transformation specification should not affect PIMs (see cell application PIM above),
modifications to transformation specifications must be accommodated in the PSM or in
transformation arguments.
Concrete
platform
PSMs related by transformation specifications must be realizable on top of a concrete
platform.
This constitutes a requirement for transformation specifications.
Transf.
arguments
7
Transformation parameters are used to introduce variations in generalized
implementation specifications; if a transformation specification is modified parameters
may be modified and new parameters may be introduced, affecting transformation
arguments.
Transf.
specification
N/A trivial
Since transformation arguments may be transformation-specific, transformation
arguments must be captured separately from PIMs so that PIMs do not become
transformation-specific. Therefore, in case of parameterization by marking, the
unmarked PIM must be kept separately from markings. The unmarked PIM and
markings can be combined into a marked model for the purposes of transformation if
necessary.
125

Design Structure Matrix. Table 7 provides an overview of the dependencies
between each of the models considered in our analysis so far. The columns of this
table correspond to the columns of tables 1 to 6. When the table is read row-wise, the
‘
7’ mark indicates that the model that names to the row is affected by the models that
name each of the columns. When the table is read column-wise, the mark shows the
models that may be affected directly as a result of a modification in the model that
names the column.
Table 7. Dependencies between models: Design Structure Matrix
Application
PIM
Abstract
platform
Application
PSM
Concrete
p
latfor
m
Transf.
arguments
Transf.
specification
Application
PIM
N/A
7
7 through
transformation
independence
is engineered
Abstract
platform
N/A independence
is engineered
Application
PSM
7 through
transformation
N/A
7
7 through
transformation
7
Concrete
platform
N/A
Transf.
arguments
7
7 through
transformation
7
N/A
7
Transf.
specification
7 7
N/A
DSMs exhibit an interesting property for our analysis: if we consider that there is a
time sequence associated with the position of the elements in the matrix, then all
marks above the diagonal are considered feedback marks [11]. Feedback marks
require iterations in the sequence of tasks executed. DSMs can be manipulated to
eliminate or reduce feedback marks, e.g., by reordering the sequence of elements in
the matrix. It is also possible to group elements of the matrix into clusters, a technique
which allows us to consider the set of elements of a cluster as a single module.
In the following section, we manipulate the DSM represented in Table 7 to show
how the dependencies between models affect the design process.
4 Dependencies between models and the design process
Preparation and execution phase concerns. Table 8 shows a reordered DSM. The
models that result from the preparation activities, namely, concrete and abstract
platforms and transformation specifications are placed in the first three positions of
the matrix. These models are grouped into a cluster, which represents the preparation
phase. A second cluster represents the execution phase, grouping application PIM,
transformation arguments and application PSM.
126

Table 8. Clustering dependencies with respect to preparation and execution activities
Concrete
p
latfor
m
Abstract
platform
Transf.
specification
Application
PIM
Transf.
arguments
Application
PSM
Concrete
platform
N/A
Abstract
platform
independence
is engineered
N/A
Transf.
specification
7 7
N/A
Application
PIM
independence
is engineered
7
N/A 7 through
transformation
Transf.
arguments
7 7 7
N/A 7 through
transformation
Application
PSM
7 7
7 through
transformation
7 through
transformation
N/A
The absence of feedback marks above the diagonal formed by the preparation and
execution phase clusters in Table 8 shows that the preparation phase does not depend
on the execution phase. This result is made possible by requirements imposed on the
preparation phase. These requirements are described in the cells of tables 1 to 6 that
correspond to the cells positioned above the diagonal formed by the two clusters.
Failure to satisfy these requirements would imply the presence of feedback
dependencies, which would require revisiting the preparation phase. The absence of
feedback marks above the diagonal formed by the preparation and execution phase
clusters can be summarized by the following design rule:
Changes in PIM, PSM or transformation arguments must be accommodated in
PIM, PSM or transformation arguments, but not in the abstract platform, concrete
platform nor transformation specification.
Table 8 also reveals the absence of feedback dependencies within the preparation
phase, since, within the cluster, no feedback marks appear above the diagonal. The
same, however, cannot be said of the execution phase: modifications in the
application PSM may affect the PIM and transformation arguments. The presence of
feedback dependencies in the execution phase is addressed through iteration in the
execution phase. An iteration in the execution phase allows a designer to gain insight
into the implications of design decisions at the PIM-level for the application PSM,
which may result in adjusting the PIM in a subsequent iteration.
However, for the design process to advance towards a stable application PIM, it is
necessary that the dependencies between PSM and PIM should eventually decrease.
Eventually, the application PIM must be such that it does not depend on design
decisions that constrain the choice of target platform. This constitutes an important
requirement for the iterative approach in the execution phase.
127

Multiple levels of models. We continue our analysis by considering the dependencies
between the models at three different levels related by transformation. Table 9 shows
the dependencies between the various models. These dependencies are clustered for
each pair of consecutive levels of models, i.e., a cluster for models of levels 1 and 2
and a cluster for models of levels 2 and 3. This DSM is build by reapplying the
transformation pattern, which explains the isomorphic nature of the dependencies in
the two clusters.
Table 9. Clustering dependencies with respect to levels of models
Abstract platform Π
1
Application PIM M
1
Transf. specification
T
1
Transf. arguments a
1
Abstract platform Π
2
Application PIM M
2
Transf. specification
T
2
Transf. arguments a
2
Concrete platform Π
3
Application PSM M
3
Abstract platform Π
1
N/A
Application PIM M
1
7
N/A
7
Transf. specification T
1
7
N/A
7
Transf. arguments a
1
7 7
N/A
7 7
Abstract platform Π
2
N/A
Application PIM M
2
7 7 7 7
N/A
7
Transf. specification T
2
7
N/A
7
Transf. arguments a
2
7 7
N/A
7 7
Concrete platform Π
3
N/A
Application PSM M
3
7 7 7 7
N/A
The table shows an overlap between the two clusters. This overlap indicates that the
design activities in the different levels are not completely independent, and that the
intermediate model PIM forms the ‘interface’ between the two clusters, as could be
expected.
5 Classifications of models
This section concludes our analysis by classifying the various models and design
decisions according to the following dimensions of separation of separation of
concerns:
– platform-independent and platform-specific concerns;
– application-independent and application-specific concerns, which correspond to
preparation and execution phases concerns, respectively; and,
– transformation-independent and transformation-specific concerns.
Figure 4 places the different models according to the first two dimensions. Three
levels of models are depicted.
128

appl i cat i on
PI M M
2
application
PSM M
3
transformation
specification T
2
transformation
ar gument s a
2
abst ract
platform Π
2
concrete
platform Π
3
application-specific application-independent
appl ication
PI M M
1
transformation
specification T
1
transformation
ar gument s a
1
abst ract
platform Π
1
pl atf orm- i ndependence
Fig. 4. Dimensions of separation of concerns and models
In Figure 4, transformation specifications are placed in the boundary between two
levels of platform-independence. This is to denote that transformation specifications
rely on the (abstract) platforms of both source and target levels of models (see Table 2
and Table 4). In addition, transformation specifications may also capture some
transformation rules which are independent of the target platform.
Similarly to transformation specifications, transformation arguments are also
placed in the boundary between two levels of platform-independence. In addition,
transformation arguments are placed in the boundary between the application-specific
and application-independent concerns area. This is to denote that arguments may be
application-specific (see Table 1), but may also capture application-independent
design decisions. Application-specific transformation parameterization is used to
improve the generality of transformation specifications with respect to specific
applications. Application-independent transformation parameterization is used to
improve flexibility of transformation specifications in general, e.g., to cope with to
variation in user requirements that are not captured in the source models but that are
to be addressed during transformation. An example of an application-independent
transformation argument determines that, irrespective of the application model, all
application parts should be allocated to the same unit of deployment of the target
platform.
In addition to the dimensions considered in Figure 4, we can also classify models
related in a transformation step as transformation-independent or transformation-
specific. This classification is relative to a transformation specification. In a
transformation step, the source application model is transformation-independent (with
respect to a transformation specification from that level of models), since it relies on
an abstract platform, which is itself transformation-independent (see Table 6). The
target application model and the transformation arguments can be classified as
transformation-specific. This can serve as a guideline to determine whether design
decisions should be captured at the source application model level or at either
transformation arguments or the target application model level.
129

6 Main conclusions and directives
From the analysis of the relations between the various models, we can conclude that:
– Feedback dependencies between execution and preparation phases can be
avoided by addressing generality requirements at the preparation phase. Failure
to address these requirements results in cycles between the execution and
preparation phases;
– Platform-independent and platform-specific models are interrelated, their
dependencies defined by transformation. The interrelation between PIMs and
PSMs is addressed through iteration in the execution phase. An iteration in the
execution phase allows a designer to gain insight into the implications of certain
design decisions at the PIM-level.
Our analysis leads to the following directives for the design process:
– Changes in PIM, PSM or transformation arguments must be accommodated in
PIM, PSM or transformation arguments, but not in the abstract platform, concrete
platform nor transformation specification.
– Dependencies between PIM and PSM are handled by iterations in the execution
phase, leading to a stable application PIM that does not depend on platform-
specific design decisions.
– Interdependent design decisions must be captured at the same level of platform-
independence. Since some design decisions are platform-specific, this imposes
constraints on the organization of models at different levels of platform-
independence. We have illustrated the consequences of interdependent design
decisions with an example in [1].
– The classification of models according to the various dimensions of concerns
serves as a guideline to determine in which models design decisions should be
captured.
References
1. Almeida, J.P.A., Dijkman, R., van Sinderen, M., Ferreira Pires, L.: On the Notion of
Abstract Platform in MDA Development. In: Proceedings Eighth IEEE International
Conference on Enterprise Distributed Object Computing (EDOC 2004). IEEE CS Press
(2004) 253–263
2. Almeida, J.P.A., van Sinderen, M., Ferreira Pires, L., Quartel, D.: A systematic approach to
platform-independent design based on the service concept. In: Proceedings Seventh IEEE
International Conference on Enterprise Distributed Object Computing (EDOC 2003). IEEE
CS Press (2003) 112–134
3. Baldwin, C.Y, Clark, K.B.: Design Rules, Volume 1, The Power of Modularity. MIT Press,
Cambridge, MA (2000)
4. Baldwin, C.Y, Clark, K.B.: Modularity in the Design of Complex Engineering Systems,
Harvard Business School Working Paper Series, No. 04-055 (2004)
5. Gardner, T., Griffin, C., Koehler, J., Hauser, R.: A review of OMG MOF 2.0: Query / Views
/ Transformations Submissions and Recommendations towards the final Standard, ad/03-08-
02, OMG (2002)
6. Gavras, A., Belaunde, M., Ferreira Pires, L., Almeida, J.P.A.: Towards an MDA-based
development methodology for distributed applications. In: Proceedings of the 1st European
130

Workshop on Model-Driven Architecture with Emphasis on Industrial Applications (MDA-
IA 2004), CTIT Technical Report TR-CTIT-04-12, University of Twente, Enschede, The
Netherlands (2004) 43–51
7. Object Management Group: MDA-Guide, V1.0.1, omg/03-06-01, OMG (2003)
8. Object Management Group: UML 2.0 Superstructure, ptc/03-08-02, OMG (2003)
9. Steward, D.V.: The Design Structure System: A Method for Managing the Design of
Complex Systems. In: IEEE Transactions on Engineering Management, Vol. 28 (1981) 71–
74
10. Warfield, J.N.: Binary Matrices in System Modeling. In: IEEE Transactions on Systems,
Man, and Cybernetics, Vol. 3 (1973) 441–449
11. Yassine, A., Braha, D.: Complex Concurrent Engineering and the Design Structure Matrix
Method. In: Concurrent Engineering, Vol. 11, No. 3, SAGE Publications (2003) 165–176
131