When ad-hoc metamodels are used, it is common to have issues related to their evolution. Managing metamodel versions and keeping the models conforming to specific (and likely old) metamodel versions is an useful alternative in some cases. However, in other situations models must always conform to the latest metamodel versions. In these cases, there is a necessity for co-evolving the models and metamodels. Moreover, in the context of Enterprise Architecture there are special characteristics that make unsuitable the application of existing Co-Evolution solutions.

To target this issue, we have developed a platform for Co-Evolution called Asimov, which is based around 3 key points:

• The evolution of a metamodel can be described in terms of a sequence of discrete steps, which give origin to intermediate metamodels.
• The co-evolution of models can be described in terms of a sequence of steps that make the model conform successively to each intermediate metamodel.
• The metamodeler knows why the metamodel evolves and has a general idea of how the model should be modied, but it is the modeler (the owner of the model) that knows exactly how the model must be modied, must approve every change, and must provide any missing information.

To support these points, Asimov offers a language to describe metamodel evolution steps, complemented by assistance steps. These assistance steps are suggestions, made by a metamodeler, about the co-evolution of the models given the corresponding evolution of the metamodels. When a model has to be evolved, the modeler loads the steps denition into the Asimov platform. This platform executes the steps one by one and presents questions about decisions that the modeler should made or about additional information that he should provide. Asimov is based on EnAr-Fusion and the issues that we encountered when we were building it led us into the Co-Creation line.

Although models of any kind can be created using standard tools and representations, specialized notations can be more e ective as they are easier to understand for domain experts. Unfortunately, adopting the previously discussed approaches, in which ad-hoc metamodels are created, makes it difficult to use representations that are not standard (i.e., are not UML class diagrams). In this line of research we have developed languages and the corresponding tool set to automatize the generation of graphical editors for ad hoc metamodels, which supports ad hoc graphical representations. In a basic sense, this tool pursues objectives similar to those of Eugenia, an editor generation tool based on Eclipse and the GMF. Moreover, our tool is much more powerful and allows the specification of richer editors with more powerful interaction mechanisms (e.g., wizards and rule-based representation selection).

Another important feature of the tool, is that it supports the metamodel composition approach previously described. This means that the description of the composition of metamodels can be used to guide the composition of their editors. This strategy fosters reusability of the metamodels, and makes much more likely the creation and adoption of ad hoc metamodels, because the costs incurred to support them are a lot lower.

Related projects:

Composing Graphical Representations of Composed Metamodels

This is one of the lines that we have barely studied up to this point, although we believe it to have an interesting potential. While in the case of EnAr-Fusion we were assuming that metamodels for EA projects can be constructed by means of composing and adapting existing metamodels, in this line we are instead assuming that metamodels are going to be dened in a kind-of organic way while the models themselves are built. This means that models and metamodels can be created at the same time, which requires special tools (editors).

Besides the issues purely related to the support of this co-creation and the maintenance of consistency, in this line of research there are other interesting topics such as those related to the usability of such tools.

In EA analysis activities, it is critical to make early statements and diagnosis from a high level of abstraction. Currently, this is dificult to perform and it requires both the involvement of experts and the elaboration of specialized artifacts. Furthermore, the complexity of these tasks increases with the size and level of detail of the models, as well as with the scope of domains covered.

Therefore, we are interested in finding approaches to support the analysis of big and complex models.

In contexts other than EA, it has been noticed that total / holistic / unfiltered visualizations may give insight about the models, providing analysts a starting point for exploration and general pattern discovery. The visualization of multidimensional data and visual scalability have been recurring problems for a long time, and there is a wide range of general purpose visualization methods that try to deal with these issues. In this context, we propose a set of domain-specific techniques that support these general methods, in order to help analysts in discovering outliers, patterns, and other anomalies over EA models.

As this research is in its early stages, key aspects that help analysis (such as interaction and view coordination) have been generally overlooked. Furthermore, important characteristics that are typical of EA models, such as imperfection and uncertainty, are currently not been taken into account. As a starting point for analysis, we suggest displaying the shape of the architecture as an holistic entity using a range of visual metaphors, to later drill down on elements of interest. From a concrete point of view, we want to give analysts a platform to provide visual support for the tasks that they should perform.

Related projects:

Visual Analysis of Enterprise Architectures

A comparative framework of query tools on enterprise architecture models

En-Ar MEnDeL