Search Results

In the automotive industry a steady increase in the number of functions driven by innovative features leads to more complex embedded systems. In the future even more functions will be implemented in the software, especially in the areas of automatic driving assistance functions, connected cars, autonomous driving, and mobility services. To satisfy the increasing performance requirements, multi- and many-core controllers are used, even in the classic automotive domains. This case study has been conducted in the steering system domain, but the results can be applied to other areas as well. Safety critical functions of classic automotive domains must fulfill strict real-time requirements to avoid malfunctions, which can potentially endanger people and the environment. For this reason, ISO 26262 requires verification of the performance and timing behavior of system critical functions.

A variety of methodologies to use embedded multicore controllers efficiently has been discussed in the last years. Several assumptions are usually made in the automotive domain, such as static assignment of tasks to the cores. This paper shows an approach for efficient task allocation depending on different system modes. An engine management system (EMS) is used as application example, and the performance improvement compared to static allocation is assessed. The paper is structured as follows: First the control algorithms for the EMS will be classified according to operating modes. The classified algorithms will be allocated to the cores, depending on the operating mode. We identify mode transition points, allowing a reliable switch without neglecting timing requirements. As a next step, it will be shown that a load distribution by mode-dependent task allocation would be better balanced than a static task allocation.

Multi-core systems are adopted quickly in the automotive domain, Proof of concepts have been implemented for power train, body and chassis, involving hard real-time constraints. However, depending on the degree of integration, it can be costly, especially in those cases where existing single-core software has to be migrated over. Furthermore, there seems to be a high level of uncertainty, whether a found solution, with regards to partitioning, mapping and orchestration of software is close to an optimum solution. Some integrated solutions demonstrate considerably less performance, for instance due to communication overhead compared to execution on single-core systems. This paper discusses a methodology, as to how to effectively and efficiently investigate the software architecture design space for multi-core software development.