Evaluation of a high performanceECU running embedded Linux

Sammanfattning: Autonomous vehicles rely on real-time embedded systems called Electronic Control Units (ECUs) and it is crucial for safety and behavioural correctness that ECUs run fast enough, in other words, within their deadlines. In this thesis a high-performance ECU used at Volvo Autonomous Solutions (V.A.S.) running PetaLinux will be evaluated. To ensure that the ECU is able to run in real-time, it will be evaluated by simulating physical models in the form of Functional Mock-up Units (FMUs) that are cross compiled to aarch64. The ECU execution time will be measured to evaluate whether the models run in real-time and to compare it with the performance of a desktop computer. With the use of profiling tools, possible bottlenecks in the utilization of ECU will be identified. The results show that the ECU is capable of running the FMU simulation models faster than real-time and that the ECU runs at less than 10% of the speed of the desktop computer. FMU communication frequency, underutilization of available instructions, unnecessary allocation and unnecessary use of synchronization primitives where identified as the key possible bottlenecks. We conclude that the ECU is likely capable of running significantly more computationally demanding models in real-time if they are optimized for the ECU.