The performance of shared-resource multi-core hardware platforms in complex cyber-physical systems (CPSs), e.g., automotive industry, can be improved using task-based parallelism through OpenMP. However, most CPS require certain level of predictability, which challenges the efficient implementation of the task-to-thread mapping process. This exploratory work build on the fact that existing mapping methods mostly use elementary or heuristic algorithms, and the idea that artificial intelligence (AI) algorithms can be used to enhance the efficiency of such processes. Accordingly, this paper (1) evaluates the suitability of AI-based techniques in improving the performance of task-to-thread mapping in the OpenMP framework, and (2) proposes a hypothesis to perform an intelligent mapping using fuzzy logic for multi-queue schedulers to improve the predictability of the system.
Read full abstract