Optimized task allocation to enhanced multi-core real-time scheduling

Abstract

To handle the increasing computational needs of onboard data processing and complicated control algorithms, modern autonomous systems require high-performance computer resources. With the rise of embedded real-time systems in a variety of industries, including as automotive, avionics, and aerospace, multi-core platforms have evolved as a compromise between performance and power efficiency. However, due to their intrinsic complexity, scheduling and schedulability analysis offer difficulties when switching from single-core to multi-core systems. As a result of the intensive research, global, partitioned and clustered scheduling were proposed. Arbitrary Processor Affinity (APA) scheduling has been proven to achieve a better schedulability for periodic tasks. APA is a clustered scheduling in which the clusters, i.e., affinity sets, are not necessarily disjoint. APA is implemented in Linux via the pull/push scheduling method, and it is also implemented in the Real-Time Executive for Multiprocessor Systems or RTEMS which is an open source Real-time Operation System (RTOS) by assigning an affinity set to a tasks. A schedulability analysis has been already proposed for APA scheduling. However, optimizing affinity sets for a set of real-time tasks such that to achieve the best scheduling is still an open question. This research project aims to provide an effective method for defining affinity sets and task assignments, with the ultimate goal of improving schedulability in multi-core real-time systems.