多核混合关键度实时系统中同步感知的调度方法研究

With the development of multicore technology and the ever-increasing performance demand of mixed-criticality (MC) real-time applications, the fundamental research on the scheduling for multicore MC real-time systems plays a crucial role in optimizing the capability of multicore processing and improving the service quality of real-time activities. This study targets the characteristics of MC tasks and the multicore resource sharing architectures, and strives for the research on the resource access protocol designs,schedulability analyses, sustainability conditions, synchronization-aware task-to-core mapping polices as well as the energy management schemes. For the problem of both priority inversion and critical inversion due to exclusive resource access in multicore resource sharing systems, we first propose the new deadlock-free resource access protocols and analyze the synchronization overheads of different blocking terms.Subject to the new protocols, this study gives the real-time schedulability analyses (e.g., response time for MC tasks and upper-bound of the system utilization), and then exploits the intrinsic relationships between the synchronization overhead and the task mapping policies. Next,we study new synchronization-cognizant approaches to assigning tasks to cores for reducing synchronizaton overheads and thus increaisng the schedulability ratio of MC tasks, and then derive the sustainability conditions to test and improve the robustness of our proposed task-to-core mapping policies. Based on these studies, we devise efficient energy management schemes to fully reclaim and feasibly allocate online slack time for better system energy savings and thus better system reliability. Finally, the proposed methods will be implemented, verified and optimized on real-time operating systems. The lauching of this research can provide new methods for multicore MC real-time applications and thus can further improve the real-time processing techniques for multicore systems.

多核技术发展以及混合关键度(MC)实时系统性能需求不断提高,研究多核MC实时调度理论,对优化多核处理能力和提高实时应用服务质量有重要价值.项目从MC实时应用本质特征及多核资源共享系统结构出发,致力于同步感知的资源访问协议、任务映射策略、能效管理机制及其可调度分析、可持续判定等调度方法研究.针对多核共享资源独占访问而导致的优先权及关键度反转问题,提出新的无死锁资源访问协议,分析并降低不同类型同步开销的上限;理论探讨该协议下响应时间、系统利用率上限等可调度条件,挖掘同步开销与映射策略的内在联系;据此研究同步感知的任务映射策略,提高任务可调度比率,并推导可持续判定条件以改善映射策略的鲁棒性;基于这些研究,设计能效管理机制,合理分配在线空闲以有效节能并维护系统稳定可靠;最终在实时操作系统上验证并优化提出的调度方法.本研究将为多核MC实时应用领域提供新的方法途径,可进一步促进多核实时处理技术的发展.

多核技术发展以及混合关键(MC)实时系统性能需求不断提高，研究多核MC实时调度理论，对优化多核处理能力和提高实时应用服务质量有重要价值。.项目从MC实时应用本质特征及多核资源共享系统结构出发，致力于同步感知的资源访问协议、任务映射策略、能效管理机制及其可调度分析、可持续判定等调度方法研究。针对多核共享资源独占访问而导致的优先权及关键度反转问题，提出了新的无死锁资源访问协议，分析并降低不同类型同步开销的上限。进而，理论探讨了该协议下响应时间、系统利用率上限等可调度条件，挖掘同步开销与映射策略的内在联系。据此，研究了同步感知的任务映射策略，提高任务可调度比率，并推导了可持续判定条件，以改善映射策略的鲁棒性。基于这些研究，设计了能效管理机制，合理分配在线空闲以有效节能并维护系统稳定可靠。最终在实时操作系统上验证并优化提出的调度方法。录用/发表学术论文14篇，其中IEEE/ACM Trans论文7篇。获得湖北省科技进步一等奖等奖励。. 本项目的研究将为多核MC实时应用领域提供新的方法途径，可进一步促进多核实时处理技术的发展。提出的部分算法策略已在操作系统内核上实现，验证了理论方法的有效性，具有较好的理论及应用价值，可为自主产权的基础系统软件研究提供借鉴。另外，提出的并行实时调度方法及技术具有良好的可扩展性，通过改进、定制相关算法，可满足实时应用需求的多样化，应用于能量效率、稳定可靠要求较高的嵌入式智能系统领域，如军工航天、智能装备、机器人、无人机、智能家居等，符合国家经济转型、保障国家安全、大力发展自主产权基础软件的战略目标。