CAAS: Cache Affinity Aware Scheduling Framework for RTEMS with Edge Computing Support

초록

Aerospace and satellite systems increasingly adopt multiprocessor architectures to support real-time missions such as autonomous flight control. As these systems shift toward edge computing with limited cache and memory capacity, efficient and predictable scheduling becomes essential. However, selecting an appropriate multiprocessor scheduler is challenging because workload characteristics and task interactions vary widely. Existing scheduling architectures?global, partitioned, and clustered?show highly variable performance depending on cache affinity and memory-access behavior, yet prior work largely focuses on improving algorithms within a single architecture rather than selecting the right one. To address this gap, we propose CAAS (Cache Affinity Aware Scheduling), a framework that characterizes workloads via reusedistance analysis and predicts the most suitable scheduling architecture based on cache-affinity profiles. Implemented on the RTEMS operating system, CAAS quantifies task-level cache affinity to automaticallydetermine whether global, partitioned, or clustered scheduling is most effective. Experimental results show that CAAS reduces execution time, improves scheduling efficiency and predictability, and simplifies scheduler configuration for emerging edge computing systems.