基于多核系统多维限制的节能实时调度研究

Energy consumption is one of the most important factors considered by many embedded real-time systems, especially wireless and portable devices. As to real-time embedded applications for the current multi-core systems, we will analyze and discuss the energy-efficient real-time scheduling strategies for multi-core systems based on Dynamic Voltage Frequency Scaling and Dynamic Power Management under the basis of the tradeoff among different architectures from simple overhead model in ideal condition to multidimensional overhead model in practical condition. In the meantime, these strategies are well related to our goals about the ratio change of energy to system performance brought by different overhead models. Then, we go deep into optimization techniques of the limited-preemptive-based energy-efficient real-time scheduling strategies in the multi-core systems, mainly for the ratio of energy to performance under the conditions of the multidimensional limitations in processor and preemption overhead constraints. Moreover, as to the aspects of sporadic DAG parallel task model or the scenario-aware dataflow graphs parallel task model, the optimization of the multidimensional-overhead-based energy-efficient scheduling strategies is explored. Finally, the preliminary energy-efficient experiments with the prototype are provided for industrial application in the field of embedded products from the levels of task management and energy management. Consequently, combining multidimensional-limitation-based energy management and real-time task model to explore the energy-efficient real-time scheduling strategies in the multi-core systems, the work is to further improve great theoretical significance about the energy-efficient design.

能耗是很多嵌入式实时系统首要考虑的因素，特别是无线移动和便携式计算设备。本课题以多核系统在实时嵌入式领域的应用需求为牵引，基于动态电压频率调节和动态功耗管理技术，在合理折中不同体系结构特点的基础上，研究和探索如何将针对多核系统的节能实时调度策略从理想环境下的简化开销模型延展到考虑更多实际约束的多维开销模型，以体现不同开销模型带来的能效比变化；并重点研究基于有限抢占调度模型的各种节能实时调度算法，尤其是研究具有处理器多维开销限制和抢占开销限制条件下的能效比评价。其次针对基于偶发有向无环图的并行任务和基于场景敏感数据流图的并行任务模型等特征，进一步开拓多维开销模型下的节能实时调度策略优化；并最终开发设计多核系统中节能实时调度器原型系统，为在嵌入式领域的工业化应用提供试验平台。因此，本课题通过研究解决如何基于多核系统多维限制实现节能实时调度的问题，将进一步完善节能设计相关理论基础。

针对现代多核系统在处理器模型、抢占调度模型、任务模型等方面存在的多维限制，基于动态电压频率调节和动态功耗管理技术，本课题通过合理折中实时性、能耗与成本等约束条件，结合实际多核嵌入式系统的节能设计要求，对节能实时调度算法的可调度性、能效和应用进行了初步研究。首先综述了多核处理器系统多维限制条件下现有节能实时调度方法；其次提出了一种开销敏感的周期任务在线多处理器节能实时调度算法，能够在保证周期任务集最优可调度性的前提下，达到实时约束与能耗节余之间的合理折中；再次提出了一种多核多处理器系统中基于松弛时间回收的开销敏感节能实时调度算法，能够利用任务实际执行所产生的动态松弛时间，降低后续任务的执行频率或关闭处理器，获得更多的能耗节余；然后设计了基于中间件的数据库同步系统，初步实现了分布式数据库系统中非线性处理的硬实时约束和能耗分布的优化；此外提出了新型的数据中心网络体系结构和通信模型，提高了网络中多核处理器系统的通信性能，降低了通信能耗；最后初步实现并验证了节能实时调度技术在多核计算机系统中的节能应用。综合上述结果，本研究为进一步完善多核系统节能实时调度理论奠定了坚实的工作基础，同时对于探索节能调度技术在服务器和嵌入式领域的大规模应用提供了重要的技术启示。项目资助发表论文8篇，SCI统计源期刊1篇，国内计算机领域一级学报1篇，EI检索国际会议3篇，中文核心期刊2篇和中国科技统计源期刊1篇。项目资助申请并授权软件著作权1项，国家发明专利1项，国家实用新型专利1项。协助培养博士研究生1名，硕士生1名，均已取得学位。项目投入经费25万元，支出9.27576万元，各项支出基本与预算相符。剩余经费15.72424万元，剩余经费计划用于本项目研究后续支出。