Architectural Modeling and Analysis of Complex Real-Time Systems

Sammanfattning: Most automation systems and other large industrial software systemshave long lifetimes, and customers expect these systems to besupported as long as they are in operation. Furthermore, softwarecomponents in these systems may be reused in different products,e.g. using a software product line approach. Hence, the lifetime ofsoftware in individual systems may be very long; several decades or even longer.Software that is used for a long time will be exposed to frequentchanges as the system evolve over time, e.g. due to adding newfunctionality, error corrections, or changing the hardwareplatform. The larger and older the system is, the harder it becomes toforesee the consequences of changes.In this thesis we present three different techniques for managing theevolution of large and complex real-time systems. The techniques arebased on analytical modeling, predicting different quality properties,e.g. temporal correctness, by analyzing a model of the software. Thefirst technique is a component model with analytical interfaces(ReFlex) that allows us to predict different properties of a componentassembly, the second is a probabilistic modeling language which isanalyzed by simulations (ART-FW), and the third technique is anextension of classical timed automata with a notion of real-time tasks(TAT).Ideally, the analytical models should evolve together with thesoftware. However, since new features are often added and theimplementation is often changed without updating the model, the modelbecomes obsolete and predictions based on the model are no longervalid. By applying the techniques proposed in this thesis, we canre-introduce analyzability; Using ReFlex we can update the analyticalaspects while re-designing the system. Unless ReFlex has been used inthe earlier design, this will require a costly redesign of thecomplete system, but consistency between the analytical model and theimplementation will be ensured. Using ART-FW or TAT theimplementation will be kept untouched by introducing a separatemodel. The drawback is that an extra effort is required to keep themodel consistent with the implementation.We have applied ART-FW in the re-engineering activity of a largeindustrial system. The results indicate that the approach is indeedapplicable on real systems.