用EDF调度实时任务和GC

Scheduling Real-time Tasks and GC by EDF

在以往对垃圾搜集器(GC)的研究论文中,为了易于分析,在GC模型中人们一般都用单调比率时限调度(RM)算法来调度GC和实时任务。近年来,为保证GC调度中硬实时任务的时限,一些研究文献提出了基于时间的GC并发调度策略。通过对算法的描述及内存需求分析,提出了用最早时限优先调度(EDF)算法来调度GC和实时任务集的方法。对用EDF和RM调度任务的延迟抖动进行了比较分析,对用EDF调度GC给出了详细的分析,而且做了相应的仿真。对比结果说明,与以RM算法为基础的其他调度策略相比,基于EDF调度的系统可以调度硬实时的非周期任务,因而更具灵活性,并且在保证硬实时任务时限要求下对系统内存的需求更小,而在任务的延迟和抖动方面,事实证明EDF调度算法并不比RM调度差。

In garbage collection（GC） models of previous work, real-time tasks and GC are mostly scheduled by rate-monotonic （RM） algorithm for simplicity. To meet the deadlines of hard real-time tasks, the time-based GC is proposed in recent years. The mutators and GC are scheduled with earliest deadline first （EDF）, the latency and jitter of EDF and RM scheduling are compared, and the detail analysis of scheduling GC by EDF is given. The simulation results are given too. The comparison results show that EDF scheduled systems can be more portable since they can schedule aperiodic tasks with hard real-time requirements and the system memory requirements can be further reduced as compared with other scheduling strategies based on RM algorithm. Furthermore, in EDF scheduled system the performance of latency and jitter of tasks are not worse than those in RM scheduled system.