异构并行系统全程序算法级低功耗优化方法研究

Reducing power consumption is the most important problem in the design and implementation of heterogeneous parallel systems. The hardware power optimization has been difficult to meet the growing demand for power optimization, so the software power optimization is widely concerned. To integrate the multiple processors computing resources, the software power optimization facing two new scientific problems, one is that the power modeling object will more complex, the other is that the energy conservation is indirectly related to optimization algorithm selection..This project proposes to carry out research on whole procedure algorithm-level on low-power optimization for heterogeneous parallel system. It intends to focus on the key technologies as follows: the formalized description of dynamic power model based on multiprocessor and multi- calculation divisions, the formalized description of communication power model based on data transmission and multitask dynamic allocation, the formalized description of static power model based on thermal analysis and real-time chip temperature management, algorithm-level on low-power optimization, as well as the verification system of the power model and optimization, etc..The primary intellectual merit of the project includes the system power model based on whole procedure，the optimal processor frequency selection and the optimal multi-calculation allocation algorithm, the integer linear programming algorithm based on profiling static analysis, the parallel task scheduling algorithm based on predictive voltage and temperature sensing. To integrate the efficacy superiority of multiprocessor resources, this project will make the power optimization technology from the hardware level and compile level to the algorithm level. This project will also provide a theoretical basis and technical support for heterogeneous parallel system low power optimization research.

降低功耗是异构并行系统设计与实现的首要问题。硬件层功耗优化已经很难满足不断增长的功耗优化需求，软件层功耗优化得到广泛关注。为融合多种处理器计算资源的效能优势，软件层功耗优化面临功耗建模对象更复杂，节能效果与功耗优化算法选择相关联两个新的科学问题。.本课题开展异构并行系统全程序算法级低功耗优化研究，拟重点研究多处理器多计算段划分的程序执行时间与动态功耗关系；研究数据传输与多任务动态分配的通信功耗形式化描述；研究热分析模型下实时芯片温度管理与静态功耗相互影响机理；研究算法级低功耗优化方法；设计并实现异构并行系统功耗优化验证原型系统。.本课题将在全程序功耗模型，处理器最优频率选择与最优下降多计算段时间分配、基于Profiling静态分析的整数线性规划、以及电压预测与温度感知的并行任务调度算法方面形成创新性成果，为功耗优化技术由硬件级、编译级走向算法级，高效应用异构并行系统提供理论依据与技术支撑。

降低功耗是异构并行系统设计与实现的首要问题。硬件层功耗优化已经很难满足不断增长的功耗优化需求，软件层功耗优化得到广泛关注。为融合多种处理器计算资源的效能优势，软件层功耗优化面临功耗建模对象更复杂，节能效果与功耗优化算法选择相关联两个新的科学问题。. 本课题开展异构并行系统全程序算法级低功耗优化研究，拟重点研究多处理器多计算段划分的程序执行时间与动态功耗关系；研究数据传输与多任务动态分配的通信功耗形式化描述；研究热分析模型下实时芯片温度管理与静态功耗相互影响机理；研究算法级低功耗优化方法；设计并实现异构并行系统功耗优化验证原型系统。. 本课题将在全程序功耗模型，处理器最优频率选择与最优下降多计算段时间分配、基于Profiling静态分析的整数线性规划、以及电压预测与温度感知的并行任务调度算法方面形成创新性成果，为功耗优化技术由硬件级、编译级走向算法级，高效应用异构并行系统提供理论依据与技术支撑。