Performance-Verifikation von Netzwerkarchitekturen eingebetteter Systeme unter Berücksichtigung von Laufzeitvariabilität

Today’s distributed embedded systems implement numerous software-based features. Those features support the user and are used in various configurations depending on the context of the system. In Automobiles, for example, different driver assistant functions can be activated with respect to the driving context. At the same time, the development of distributed embedded systems is characterized by high cost pressure. This has a strong impact on the development process. The better the needed resources can be predicted the less are the production costs. Simultaneously, distributed embedded systems underlie real-time constraints, especially for communication. The adherence of those requirements to the network is ensured during the network performance-verification. The goal is, to fulfill those requirements with the lowest possible amount of hardware resources. Herein lies the challenge of this process step. The complexity rising through the runtime variability of the system complicates an exact performance-verification and thus, the finding of a cost-efficient network architecture. In the work at hand, a new approach is presented aiming to solve this challenge. This approach foresees two variants. Both variants enable the performance-verification of current distributed embedded systems by considering runtime variability on the basis of an analysis model and a network performance model. The analysis model describes runtime variability, software- and network architecture as well as their interrelations. The network performance model specifies the performance of the network. While the focus of the first variant lies on a preferably exact verification, the second variant puts more emphasis on the scalability regarding runtime variability. An integrated network performance model based on the first variant has been constructed in order to receive a scalable solution. This model describes runtime variability within the network performance model. For this solution, new heuristics have been designed that – compared to the first variant – reduce the computing time significantly in combination with a mixed-integer linear program (MILP)-solver. The presented approach has been evaluated by the example of automotive e/e architectures. The results show that the approach fulfills the requirements of current distributed embedded systems and that it can enable significant improvements in resource-efficiency. Within a case-study, it has been shown that improvements of 95 percent in average are possible compared to a static performance verification. It has further been shown that the second variant scales for systems with a high degree of runtime variability. Furthermore, multiple factors influencing the potential of improvement of the presented approach have been identified