Dynamic preemption threshold scheduling for specific real-time control systems

Abstract

Application specific operating systems (ASOS) are developing quickly as a new trend in real-time control systems development. It often belongs to system on chip. The scheduling for ASOS should satisfy two basic demands (a) context switching overheads are not significant; (b) the scheduling should use small amount of RAM memory. According to characteristics of ASOS, we present a novel scheduling algorithm, named dynamic preemption threshold (DPT) scheduling, which integrates preemption threshold scheduling into the EDF (earliest deadline first). The scheduling can achieve greater processor utilization, theoretically even up to all of a processor capacity. Meanwhile, the preemption times between tasks can be effectively decreased using DPT scheduling by two ways: 1) threads allocating; 2) dynamic thresholds regularly adjusting at runtime. With the reduction of task preemptions, memory requirements are also decreased. In addition, the DPT gives an approach to transform a static model to dynamic model seamlessly. The DPT algorithm can perfectly schedule a mixed task set with preemptive and non-preemptive tasks, and subsumes both as special cases. Thus it remains the scheduling flexibility and also decreases unnecessary context switching and memory requirements at runtime.