Publications

Timing behavior predictability is a necessary real-time requirement for future switched Ethernet-based networks in the automotive and similar domains. Schedulability of such safety- and time-critical networks has to be guaranteed. The paper proposes a system architecture for future Ethernet-based advanced driver assistance systems (ADAS) and analyzes its timing requirements on transmission and forwarding deadlines. It studies the schedulability for strict priority combined with earliest deadline first (EDF) scheduling. For that, conditions on bandwidth utilization factors guaranteeing schedulability in the whole network are formally determined.

The development of complex systems often leads to delays in integration and roll-out, massive cost overruns, and in many cases to sub-optimal solutions regarding system performance and robustness. In order to overcome these issues, model-based systems engineering methodologies can be employed. The application of simulation models in addition to system specifications increase the engineers' understanding of complex system behavior. The approach also allows the detection of incomplete or inconsistent functional requirements as well as other design failures. With increasing complexity of system models it is necessary to optimize parameters to achieve a resource-efficient system. Indirect simulation is a well-known tool for this task, and there are numerous heuristics that can control the search in the parameter space. However, this task is much harder if the design parameters are not only numerical values from a predefined interval. In many cases the design issue (and optimization task) involves decisions about how a system architecture should be composed out of modules and interfaces / connections. In such cases, already the generation of feasible design variants may be hard to automate as a part of an optimization algorithm. Domainspecific knowledge and technical constraints have to be taken into account, and the non-numerical parameters have to be encoded such that they become accessible by the controlling heuristic. This paper presents a model-based methodology to optimize the architecture of a wireless sensor network that is designed for future avionic applications. We show how dynamically reconfigurable simulation models can be used to simplify an optimization of architecture or topology of a system. The system modeling and design tool MLDesigner is used for model implementation and simulations. In addition to that, the heuristic optimization process is described in such a model, thus controlling on a meta level the actual optimization. It can be changed or adapted easily, because it is not programmed but described with a high-level model. Index Terms: Performance evaluation, global optimization, discrete event simulation, probabilistic metaheuristic, simulated annealing, model reconfiguration