Having discussed the origin and the determinants of the multiple viewpoints on a semantical and mereological level, this section embraces on the syntactic and formalized specification of viewpoints. Hence, the relationships between viewpoints and meta models will be discussed in the following. In section 2.2.1 the similarities of multi-view modeling to relational databases has been discussed. It is therefore no surprise, that there are also similarities in the specification of the syntactic viewpoint relationships. “Integrating heterogeneous metamodels and instances bears similarities to the well-known problem of schema integration of heterogeneous databases ((REDDY ET AL., 1994; SHETH ANDLARSON, 1990))“ (BURGER, 2013, p. 1:4).
In order to specify different relationship classes, a generic architecture framework is first introduced in the following. SINZ (1997, 2002) introduced a generic architecture framework for information systems, feasible for establishing a generic and comprehensive classification scheme referred to later on. The goal of this framework is to enable the analysis and specifica- tion of complex information systems architectures. Figure 27 illustrates the components of the generic architecture framework. The framework utilizes a layered approach for handling the complexity of information systems that is introduced in the following:
• Definition of model levels, structuring the model system (cf. section 2.1.1). Every model level describes the information system comprehensively, thereby taking a certain per- spective on the information system. The different perspectives are aligned to the different goals pursued during the modeling process. Common perspectives, are the differentiation according to task and resource, or according to an inside and outside perspective (cf. sec- tion 7.1 and Figure 41 for the application of these perspectives in the SOM method). The construction rules for each model level are defined by a corresponding meta model.
Model Structure
Views
Meta Model
Pattern
MM
iMM
j View1 View2 Viewn View1 View2 .... P1 P2 Pm P1 P2 ....RMM
ij Model Level Relation Between Model Levels Model LevelModel Level
Hierarchy
RP1 RP2 Relation Between ViewsFigure 27: Generic architecture framework (SINZ, 2002, p. 876)
• Definition of views for each model level. Views are specified by defining a projection operator based on the meta model of the corresponding model level. Each view de- scribes usually a partition of the model. All views are integrated through the common meta model. The combination of the views on one model level gives the comprehensive model on that level. Common views, according to Sinz are e.g., data view, function view, structure view, or behavior view (cf. section 7.1 for the application of the structural and behavioral view in SOM business process modeling).
• Definition of relationships between model levels. For each pairwise relation between model levels, relations are defined by specifying a relationship meta model (RMM). A RM Mij relates meta model elements of meta model i (M Mi) to meta model elements of meta model j (M Mj). The specification of these relationships between the meta mod- els of different model levels integrates the comprehensive model system. Furthermore, assignments and transformations between model levels can be defined (SINZ, 1997, p. 6). • Definition of patterns (P) and relationship patterns (RP) for each model level and be- tween model levels, respectively. Patterns are used to limit the valid structures of a model level, e.g., based on heuristic modeling experience or as a means of structural integrity constraints (SINZ, 1997, p. 4).
The existence of the relationship meta models (RMM) does not only contribute to an inte- gration of the meta model, but also to the definition of relationships between the views. These
relationships are transitively defined by the relationship between the corresponding meta mod- els (SINZ, 2002). However, Sinz already stated that “In many modeling languages only more or less isolated sub-meta models for the individual views are given. By contrast, an integrated meta model is not available. From a methodological point of view, this represents a serious deficit, since the alignment of the individual views is prevented or at least hindered“24 (SINZ, 2002, p. 877).
The positive effects of an integrated meta model are manifold. The most important ones are described briefly: The definition of an Integrated Meta Model (IMM) defines relationships not only between the meta models but also on the views, derived from the meta models. Conse- quently, syntactic dependencies can be defined once on meta level and are effective for any created instance. Adding a new meta model to the model system, and therefore to the IMM, enables immediate transformation and alignment between all other meta model instances and the new one. Moreover, the definition of an IMM facilitates a general understanding and doc- umentation of the complex multi-view architecture. Zachman already stated for information systems architectures in 1987, “since the technology permits “distributing“ large amounts of computing facilities in small packages to remote locations, some kind of structure (or architec- ture) is imperative because decentralization without structure is chaos“ (ZACHMAN, 1987, p. 454). The same holds for the decomposition of the model system into model levels and views. As the integration of meta models plays an important role, the different approaches for meta model integration in the context of integrated multi-view modeling are now discussed.
5.3.1 A Dichotomy of Integrated Multi-View Modeling
In case of multi-view modeling, the tight coupling of the information systems architecture pre- sented before is not feasible. By contrast, more flexible approaches to integrate multiple views, i.e., models, need to be considered. In the following, a dichotomy of integrated multi-view modeling is established.
5.3.1.1 Meta Model Integration
Figure 28 illustrates the case of an already existing integrated meta model. This case refers to the generic architecture framework for information systems introduced in section 5.3. The object system, i.e., the relevant part of the real world, is mapped into a model system (cf. sec- tion 2.1.1) during the modeling process. Integrity of the model system, and the multiple views derived, is established by the integrated meta model. Views on the model system are defined as projections, that can be applied to the integrated meta model. The integrity of the views is therefore transitively realized by the the design of the meta models.
Model System
Integrated
Metamodel
View 1
View 3
View 2
Object
System
Viewpoint 1 Viewpoint 2 Viewpoint 3Figure 28: Meta model integrated multi-view modeling (cf. BORK ANDSINZ(2011b))
5.3.1.2 Viewpoint Integration
As already stated by SINZ(2002), the common case is that, where method engineers are forced to integrate “more or less isolated sub-meta models“. Hence, each viewpoint has its own meta model that needs to be integrated towards an integrated multi-view modeling method. Figure 29 illustrates this case. The object system is mapped into several views whereas each view is constructed according to a viewpoint. The viewpoint specification itself is based on a specific meta model. The model system is therefore not integrated by the meta model. The integration of the model system, by means of the multiple views visualizing it, has to be realized by a previous integration of the corresponding viewpoints.
Object System View 1 View 3 View 2 Viewpoint 1 Viewpoint 2 Viewpoint 3 Metamodel 1 Metamodel 2 Metamodel 3
5.3.2 Discussion
There are different ways of integrating multiple views. The presented dichotomy compared meta model integrated and viewpoint integrated approaches. The latter enables more flexibility, however, at the cost of specification effort. Whenever a new viewpoint is introduced it has to be analyzed and probably integrated with all other viewpoints, whereas in the former, it only has to be integrated with the integrated meta model. Both approaches can be aligned to the categories identified by the IEEE: “There are two common approaches to the construction of views: the synthetic approach and the projective approach. In the synthetic approach, an archi- tect constructs views of the system-of-interest and integrates these views within an architecture description using model correspondences. In the projective approach, an architect derives each view through some routine, possibly mechanical, procedure of extraction from an underlying repository“(IEEE, 2011, p. 22).
Boulanger et al. contributed to this understanding by stating: “The first and most common one is to consider that views are projections of a hybrid reference model. In this approach, the reference model aggregates all the information about the system. Views are queries on the reference model; they perform projections hiding irrelevant information when studying a par- ticular aspect of the system. The second approach uses views as partial definitions or expected observations of the system“ (BOULANGER ET AL., 2010, p. 313).
Albeit all the benefits of the projective approach (i.e., the meta model integrated approach), it must be emphasized, that the relevant literature not considers the actual creation and processing of the models. The authors assume, that with this multi-viewing approach all integrity checks and inconsistencies are solved by design. However, while developing multi-view modeling tools this assumption does not hold. In the following, two approaches are introduced that emphasize on the interactions between modelers and multi-view modeling tools - the way of carrying out multi-view modeling.