Högskolan i Skövde

A basic premise of model-driven development (MDD) is to capture all important design information in a set of formal or semiformal models, which are then automatically kept consistent by tools. The concept, however, is still relatively immature and there is little by way of impirically validated guidelines. In this paper, we report on the use of MDD on a significant real-world project over several years. Our research found the MDD approach to be deficient in terms of modeling architectural design rules. Furthermore, the current body of literature does not offer a satisfactory solution as to how architectural design rules should be modeled. As a result developers have to rely on time-consuming and error-prone manual practices to keep a system consistent with its architecture. To realize the full benefits of MDD, it is important to find ways of formalizing architectural design rules, which then allow automatic enforcement of the architecture on the system model. Without this, architectural enforcement will remain a bottleneck in large MDD projects.

Current techniques for modeling software architecture do not provide sufficient support for modeling architectural design rules. This is a problem in the context of model-driven development in which it is assumed that major design artifacts are represented as formal or semi-formal models. This article addresses this problem by presenting an approach to modeling architectural design rules in UML at the abstraction level of the meaning of the rules. The high abstraction level and the use of UML makes the rules both amenable to automation and easy to understand for both architects and developers, which is crucial to deployment in an organization. To provide a proof-of-concept, a tool was developed that validates a system model against the architectural rules in a separate UML model. To demonstrate the feasibility of the approach, the architectural design rules of an existing live industrial-strength system were modeled according to the approach.

An important purpose of architectural design is to ensure that the system meets its quality requirements by defining a set of system wide design decisions. An important part of these design decisions is the set of architectural design rules that shall be followed by developers in the detailed design. The state of practice is to define these rules in natural language and to use manual reviews to enforce them. This way of transferring the rules to the developers is however error prone and requires a lot of effort from the architects since natural language is ambiguous and open for different interpretations and rule following have to be checked with manual reviews. This paper reports from an action case study where a novel approach for architectural modeling and automated conformance checking has been investigated regarding its ability to better communicate architectural design decisions to the developers. The findings indicate that the novel approach is significantly more effective than the state of practice. The findings also show that an important reason for this is that using a tool for conformance checking allows the developers to learn the rules by experimenting.