Making the Business Case for Software Factories
Gunther Lenz, a Microsoft Architect, and Christoph Wienands, a Software Engineer working at Siemens undertook the project to evaluate the benefits of establishing a Software Factory based on Microsoft Visual Studio 2005 Domain Specific Languages (DSL) tools.
While it seems obvious that increased reuse, higher level of abstraction, and increased automation provide advantages to today's text-based software-development efforts it is not enough to justify the necessary upfront investment to upper management and other stakeholders. Therefore, we started to look for quantitative data that would clearly show the business case for using this paradigm.
The goal of this analysis was to identify automation and reuse opportunities in the areas of framework and application development. To that effect, the authors developed a Software Factory Template.
The implementation of the Software Factory Template is based on the Software Factory Schema. In our case, we provide multiple T4 templates—for artifact generation, wizards, and a DSL—that are bundled into installer packages to provide the developer with additional tools and context-sensitive automation mechanisms.
Their project has two components:
- Generation of some framework C# artifacts using a DSL
- Support of application developers through guidance and automation
In this first study, they have found that:
For the developer, instead of entering 320 lines of code per definition and use, only a few strings have to be entered, and some decisions that are based on valid options have to be made. The net result is not only time savings, but also quality increase by providing only valid choices and the ability for validation of input before artifact generation.
Overall, this translate into:
The DSL development increased productivity by a factor of 10 with a ROI of more than 100 percent.
For the guidance package, we could show a ROI of more than 300 percent with minimal investment.
The authors plan to tackle more complex questions in the future:
- How does this approach scale to broader domains?
- How does the tool support work for more complex implementations?
- How much savings will there be on maintenance efforts?
- How do we handle evolution of DSL meta-models and models?