From small and very specialized applications, real-time embedded systems are now evolving towards large distributed and interconnected systems. The construction of such systems is becoming increasingly complex, while being put under heavy pressures (economic, mission criticality, time, etc.). We see a current trend to extend and reuse existing specification and modeling techniques for embedded systems under the "Model Driven Architecture" approach (MDA). Expected benefits range from a better control over the application life-cycle to the integration of performance, analysis or verification tools. In this paper, we take a very pragmatic position and illustrate how building Distributed Real-Time systems (DRE) for the High-Integrity domain in a Model Driven Development (MDD) process may fail to address core requirements, and why going "back to the basics" of the code and implementation is required to avoid missing the strongest requirements; and avoid a situation in which the MDD fails to deliver its promises. Our claim is that MDD provides value to the engineering of complex system, if and only if it can take full advantage of the expressive power of the models to help the user in certifying or validating its system. This includes full control of the code generation, validation and verification or testing process. In the following, we show some limits in current MDD-based DRE projects. We discuss how a careful use of a modeling language like AADL can reduce them, by separating modeling concerns from the underlying execution environment. We then introduce our work in this domain, demonstrating how both a unified modeling approach, combined with precise code generators can provide the user full control and confidence when building its own DRE systems.
展开▼
