Real-Time Computing on High-Performance Multi-Core Platforms

Embedded systems are at the heart of the ongoing digitalisation that transforms all areas of society. An increased need for higher levels of autonomy can be observed in several domains, such as automotive, avionics, and robotics. To deliver the necessary processing capabilities, high-performance Multiprocessor Systems-on-Chip (MPSoC) platforms are considered to host safety-critical embedded applications subject to timing constraints. However, commercial MPSoC platforms are not designed for real-time workloads and are inherently unpredictable due to resources shared between compute cores, such as the memory hierarchy. Software-based techniques, like memory-centric scheduling and phased execution models, have been proposed to isolate the execution of a task from interference caused by other cores. At the same time, there has been a shift from classical static Real-Time Operating Systems (RTOS) to POSIX-based RTOS, which host several applications simultaneously. This talk will provide an overview of the area of real-time systems before focusing on software-based techniques to host real-time workloads on commercial MPSoC. We present techniques to increase the efficiency of phased execution models on platforms with scratchpad memory. We also discuss how timing guarantees can be provided to applications that utilise a commercial POSIX-based RTOS, considering resource reservations and distributed systems.