Write once, deploy N: a performance oriented MDA case study

Optimizing the performance of distributed database applications is difficult to combine with middleware vendor independence since cache, transaction and cluster configuration is database and application server specific. The most promising approach is to use a flexible code generator. The model driven architecture can become a successful standard for model based code generation by offering a set of high quality code generation languages. To focus the comparison of such languages on criteria that matter in practical development, there is an urgent need for more, and more realistic, case studies. Therefore, we present a complex middleware performance pattern that should be generated for at least two application servers. The code generator should have maximal reuse, separation of concerns and evolution support. From this case study, we derive the requirements for model refinement and code generation languages. We illustrate how existing UML standards can be integrated to satisfy these requirements and conclude with an overview of the language features that were needed to provide an elegant solution to the case study.