Open Access

Julian Kienberger

• In the past decades, the automotive industry has been seeking to include more and more features in its vehicles while simultaneously attempting to reduce the number of “Electronic Control Units” (ECUs) that execute the corresponding embedded software. As technical and economic limitations prevent furthermore rising the single-core computing power, the migration of ECU software to target platforms featuring multiple “Independent Execution Units” (IEUs), e.g., processor cores, is enforced and the necessary paradigm shift towards multi-core technology is in full swing. In order to eventually exploit the extra processing power, there is much additional effort needed for coping with the tremendously increased complexity of such systems. This is largely due to the elaborate parallelization process consisting of partitioning, mapping and scheduling software parts as tasks on different IEUs. The associated, rapidly growing number of possible solutions results in a combinatorial explosionIn the past decades, the automotive industry has been seeking to include more and more features in its vehicles while simultaneously attempting to reduce the number of “Electronic Control Units” (ECUs) that execute the corresponding embedded software. As technical and economic limitations prevent furthermore rising the single-core computing power, the migration of ECU software to target platforms featuring multiple “Independent Execution Units” (IEUs), e.g., processor cores, is enforced and the necessary paradigm shift towards multi-core technology is in full swing. In order to eventually exploit the extra processing power, there is much additional effort needed for coping with the tremendously increased complexity of such systems. This is largely due to the elaborate parallelization process consisting of partitioning, mapping and scheduling software parts as tasks on different IEUs. The associated, rapidly growing number of possible solutions results in a combinatorial explosion and thus spans a vast search space. Mastering this challenge requires both the prevention of impending data inconsistencies caused by race conditions and the avoidance of additionally required synchronization effort originating from the distributed execution of an application across different IEUs. Therefore, there is a strong need for innovative concepts, methods and a continuous tool chain to, on the one hand, support the migration of legacy software by efficiently parallelizing it (i.e. first partition and then distribute the obtained parts to different IEUs) and, on the other hand, ease the creation of embedded multiple-IEU applications from scratch. In addition, it is crucial to support the development process of software being compliant with the leading standard “AUTomotive Open System ARchitecture” (AUTOSAR). This PhD thesis starts with a description of the circumstances and factors leading to this situation before stating correlating challenges and deriving concrete objectives to eventually cope with them. In addition, a demonstrative running example is introduced that bases on a car’s wheel speed measurement function and is intended to illustrate ideas and methods throughout the thesis. Afterwards, basic knowledge, applied concepts and techniques are covered in order to provide a sound basis for the later presented methodology. Prior to the main chapter, relevant existing research approaches are introduced, categorized and evaluated with regard to their significance for the suggested methodology and conducted case studies. The pivotal notion of the presented approach is to utilize a tool-aided data dependency analysis performed on AUTOSAR system descriptions (referred to as “models”) to eventually determine advantageous partitions as well as initial task-to-IEUs mappings. Therefor, the analysis’ results are used for a verification and validation process that modifies the model to achieve “multiple-IEU robustness”. This robustness is universally valid – irrespective of a specific partitioning and mapping solution. Based on this and following the overarching methodology’s concepts, concrete partitioning strategies and algorithms are applied in order to efficiently identify advantageous disjoint subsets even within highly complex models. The succeeding mapping step then distributes the obtained model parts to available IEUs in an expedient manner. Afterwards, the determined solutions serve as input for the simulation within a timing tool for embedded multi-core systems. Here, this initial solution is evaluated with respect to scheduling quality (e.g. fulfillment of task deadlines) and metrics like cross-IEU communication rate, communication latencies, memory consumption or load distribution. Lastly, a subsequent optimization process enhances the initial solution, iteratively generates improved variants and enables a comparative assessment. In order to demonstrate the realization as well as to prove practicability and benefit, three case studies are presented that illustrate the feasibility and usage of the methodology by means of real-world examples coming directly from the industry. The added value is shown with the respective evaluations, e.g., via reduced effort and increased solution quality achieved with sticking to the introduced concepts and methods. In conclusion, the proceeding is summarized and originally formulated objectives are confronted with the reached achievements. Finally, an outlook on currently prevalent trends within the automotive industry is given just before the concluding depiction of impending tasks as well as promising concepts for further development.…