Partitioned Cache Aware Dynamic Scheduling for Real-Time Applications on Multicore Processors

Abstract

Efficient task partitioning and scheduling on multicore processors are critical for optimizing performance and resource utilization in real-time systems. This paper explores a dynamic approach to task partitioning and scheduling, leveraging Intel Cache Allocation Technology (CAT) and pseudo-locking to enhance predictability and reduce inter-core interference. By dynamically allocating cache resources to critical tasks, partitioning high-frequency tasks into a separate cluster and isolating them from contention, the system achieves improved schedulability. Additionally, an adaptive Earliest Deadline First (EDF) scheduling algorithm is introduced, which allocates the tasks to free cores in real time based on workload variations and resource availability. The proposed techniques are validated through typical applications in signal processing and other similar systems, where high throughput, low latency, and strict timing constraints are paramount. Experimental results of the Modified-EDF approach demonstrated a reduction of 4.6 % in Worst-Case Execution Time (WCET) compared to SCHED_FIFO and a decrease of 2.3 % in CPU utilization Similarly, it achieved a 4.2 % improvement in WCET over SCHED_RR and a 2.3 % improvement over SCHED_DEADLINE., highlighting its efficiency gains through deadline sensitivity and cache-awareness, thus making this approach highly suitable for safety-critical and high-performance computing environments.