Why MDD Works (Part 1)?

When I started Intelliun in 1999, I wanted to develop a radically new programming paradigm that allows a large number of developers to concurrently and independently build small applications. These applications can be quickly assembled to solve large and complex problems including business/enterprise, scientific and even mobile applications. In a matter of fact, the name Intelliun stood for INTELLIgent UNiverse, which is the environment to run intellets (i.e. small units of intelligence).

I remember I was reading "Crossing the Chasm" by Geoffrey A. Moore, and one of the things it warns against is starting a business that introduces a paradigm shift. I did agree with the rational, and I had another idea I also wanted to develop anyway. That idea was MDD (didn't use the term at the time though). I figured it would be much easier to promote and gain traction with MDD, but to my surprise, that was not the case.

Around the middle of 2000, at JavaOne, we introduced VE 2.0 beta, our first public release of The Virtual Enterprise. The conference was a blast! We got a lot of interest and intrigue and generated many leads…after all we were still at the height of the dot com bubble. Following up on those leads, however, was a different story.

Managers and business people loved us. We are demonstrating a technology that will radically cut their development time and cost, and help reduce their pain of hiring more highly scarce developers. Even ones that came up through the technical ladder were very impressed. I remember flying to NY with our VP of Sales and meeting with a director at Lockheed Martin that we were introduced to by a visitor to our booth at JavaOne. He brought in his chief architect, who was an older gentleman with many years of experience and the gray hair to prove it. I still remember the look on his face when he walked into the room.

Here this little kid that is going to tell us how to develop software better. After all, he has seen it all, from punch cards to CASE tools to this new Java thing. Nevertheless, he patiently sat down and went through the presentation and the demo. At the end of the demo, he was so impressed that he stood up and shook my hand with both hands. He saw enough to accept that it can be done and that it was different.

Developers, on the other hand, hated us. We got significant skepticism and push back, some because of job security and fear of change, but the majority was out of the lack of understanding. It seemed to them like a major leap from how they're developing applications, and even if it worked, it will only be able to handle simple applications.

What I would like to do in the next couple of entries is to explain MDD from my point of view and demonstrate how it is really an incremental improvement and not a radical departure on current software development practices.

Scaling Model Driven Development : Introduction

Demos for software tools usually involve some sort of toy project so you can get up to speed with the environment and concepts without having to worry about the details of a complex application. Unfortunately, it's exactly those complex details that rise up to bite you when you adopt a new tool or process and try to use it on a real-world project.

I'm going to do a series of postings about the stuff you don't always notice in the demos, and how it can make the difference between a big mess and a clean scalable model.

You know those scotch-taped-together wall-sized database diagrams with hundreds of tables and an impenetrable maze of relationships between them? Don't do that. There's nothing that says all your classes have to be on the same canvas. You can divide up analysis-level classes into packages exactly the same way you'd divide your Java implementation classes into packages. The familiar principle of reducing coupling between packages applies to models.

A totally standalone model has very low coupling but is otherwise pretty useless. After you divide up your model, you need to hook the pieces back together somehow. But if you just add back in direct relationships then you've simplified the visual design but not the underlying complexity.

In the next post in this series I'll talk about the kinds of coupling that should be allowed between pieces of the model, and the specific modeling artifacts that allow you to incrementally add complexity without breaking previous work. Hopefully it will all lead up to some posts of how to write reusable models, but it depends on how ambitious I'm feeling.