多核混合关键度系统中软件级节能关键技术研究

With the increasing requirements for size, weight, and power of Cyber-Physical Systems (CPS), all kinds of tasks are being integrated into a single platform and become mixed-criticality systems. This project will research the key technologies for software-level energy savings in mixed-criticality systems on multicore platforms. This project includes the following contents. First, it will build the collaborative energy-saving framework in order to provide support for task mapping, slack time management, and runtime controlling. Second, it will present the task-mapping method which incorporating memory with processor cores. This method can provide support for improving the throughput of non-safety-critical tasks and reducing the energy consumption of processors and memory while guaranteeing the realtime requirements of safety-critical tasks. Third, it will research the slack time management method for intra-cores and inter-cores such as slack-reclaiming, slack-prefetching, and slack-reserving, etc. This method will provide support for controlling of tasks and memory. Fourth, it will explore the runtime energy-saving method for task state-controlling and memory power-controlling, which will improve the utilization of processors, the throughput of non-safety-critical tasks and reduce the energy consumption while meeting the deadline requirements of safety-critical tasks. This project will provide theoretical achievements, algorithms, and implementation techniques for the design of energy-efficient mixed-criticality systems on multicore platforms, and help to improve the reliability, throughput and save energy consumption of mixed-critically systems.

随着信息物理融合系统在体积、重量和功耗等方面要求越来越高，不同类型的任务正日益集成到单一平台中从而形成混合关键度系统。本项目以多核混合关键度系统为研究对象，研究软件级节能关键技术。本研究包括建立多核系统协同节能框架，为任务映射、空闲时间管理和运行时控制提供支持；提出内存和处理器核相结合的任务映射方法，从而在保证安全关键任务实时性的同时，提高非安全关键任务吞吐量并为降低处理器和内存的总体能耗提供支持；研究运行时核间和核内空闲时间回收、预取和预留等空闲时间管理方法，为任务和内存控制提供支持；以及探索运行时任务状态控制和内存功耗控制相结合的节能方法，从而在保证安全关键任务满足截止期限的情况下，重用空闲时间，提高处理器的利用率、非安全关键任务的吞吐量和降低能耗。本研究为设计多核平台中能量有效的混合关键度系统提供理论依据、算法和实现技术，有助于提高混合关键度系统的可靠性、吞吐量和节省能耗。

混合关键度系统的节能在保证系统正确性的同时，对于提高系统的CPU和内存资源利用率并降低能耗具有重要的意义。本项目对基于DVFS和DPM的多核系统协同节能框架、内存映射和处理器核映射相结合的任务初始化方法、核间和核内空闲时间管理方法、以及任务运行状态控制与内存功耗控制相结合的节能方法进行了研究。在协同节能框架、任务在CPU核上的驻留位置分配、参数分配，和运行时的迁移和回归、运行时空闲时间回收、预取和重用、以及内存和CPU的协同功耗优化方面取得了一系列研究成果。.截至2019年12月31日，项目组在国内外期刊和学术会议上发表的标明基金资助号的论文20篇，其中SCI收录9篇，EI收录7篇；IEEE Transactions及JSA等国际一流刊物上发表论文7篇，国际一流会议IEEE INFOCOM/SECON上发表论文2篇；申请国家发明专利7项，其中授权2项；获得软件著作权9项。1名教师晋升为副研究员，1名博士晋升为副教授；培养博士毕业生4名，培养硕士毕业生8人。参加嵌入式系统相关领域境外召开高水平学术会议6次，国内召开学术会议10次，与国内外多所大学和研究机构进行了学术交流。.本研究有助于解决混合关键度系统在能量受限和安全关键的领域，例如移动医疗和移动安全检测等领域应用中的技术限制，具有较强的理论价值和应用前景。