Agile MDA: A Roadmap for Sophisticated Modeling Tools

I’ve been pretty harsh with respect to the Model Driven Architecture (MDA) in previous writings. I still question the viability of the MDA for most organizations, although respect the fact that many people are still going to attempt it within their organization. This begs the question “how do you take an agile approach to the model-driven architecture (MDA)”. Figure 1 summarizes what I believe to be a realistic “agile MDA” approach which applies the concepts of Agile Model Driven Development (AMDD) to the MDA. It’s important to note that you need to look beyond MDA-based modeling tools to make MDA work. Figure 1 depicts one of three basic AMDD approaches, the most complex one in this case.

Figure 1. An AMDD approach to MDA.

Below are some point-form thoughts which I will flesh out in the coming weeks.

Critical Success Factors:

You should apply the principles and practices of Agile Modeling (AM)
The primary goal is to produce working software which meets the needs of your stakeholders otherwise you’re not really doing MDA.
Active stakeholder participation is crucial to your success, so you need to find ways to work with your stakeholders effectively (see below).
Testing is a critical issue throughout the entire development effort. Acceptance tests should be included in your PIMs and technically-oriented tests in your PSMs.
Take an evolutionary (iterative and incremental) approach. Serial/traditional MDA isn’t a very good option as the focus is often on the models/documents and not on the working software.
Developers need to learn fairly sophisticated skills. People/communication skills are critical, luckily agile developers are focused on these skills, but so are modeling skills (which few developers seem to focus on).
You’ll need a wide range of model to make this work. Figure 2 maps the MDA model types to various modeling artifacts. You don’t need to create all of these artifacts, but you will need some.

Figure 2. Artifacts for Agile MDA.

Working with stakeholders:

To work effectively with stakeholders, you need inclusive modeling techniques as Figure 1 indicated. These techniques are simple and use simple tools, see AgileModeling.com/artifacts/ for a list, as well as Figure 2.
Inclusive techniques foster Active Stakeholder Participation.
Stakeholders don’t understand complex diagrams, something we learned in the 1980s.
Figure 2 includes inclusive modeling as part of the overall MDA vision (I’ve had to add this concept to the overall MDA picture).

Thoughts about UML & MDA:

If you want to do the full tool-based MDA thing, you should use the inclusive techniqiues to gather info then capture the information in your tools. Not ideal, but at least you have a bridge to your stakeholders.
The MDA seems to be missing the concept of inclusive modeling.
The transformation between inclusive models and PIMs is manual, requiring skill on the part of the modeler.
The PIMs and PSMs require a sophisticated modeling tool, few are out there. Bridgepoint, TogetherCC, and OptimalJ are leading candidates.
The UML models seem to be oriented towards PSMs, not PIMs (hence you need to extend the UML to really support MDA, something we see the tool vendors do all the time).
You’ll likely need to go beyond UML even if you do take a visual approach.
Figure 3 maps the model categories to various modeling artifacts, indicating which techniques are inclusive, which are complex, and which are “in between”.
We need to extend the UML beyond object and component technology.

Thoughts about Tools:

I prefer to use simple tools for “up front” requirements/analysis activities, more complicated tools (e.g. CASE) for design.
For MDA to work, PIMs (the up front models) need to be captured in the tool.
You’ll still need to code, either in a high-level action language from which program code is generated, or program code itself. Depends on your tool choices. Pick the approach that works best for you.
Many tools seem to claim to be MDA, few really seem to achieve this goal.
There is far more to succeeding at the MDA than just tools.
It’s doubtful that the tools will work together effectively, integration through XMI will fail in practice, so pick your poison and move forward.

Conclusion:

MDA is still going to be hard — you need the modeling skills, you’ll become tied to your tool vendor, and so on (see the list of challenges).
Most developers prefer text-based, not visual, approaches to “modeling” in my experience. This makes me question a diagram-based approach for many shops.

Figure 3. Techniques for each modeling category.