支持绿色异构计算的能效感知调度模型与算法

发布时间：2019-08-03 10:49

【摘要】：异构调度可使大规模计算系统采用并行方式聚合广域分布的各种资源以提高性能.传统调度目标追求高性能而忽视高效能,远不能适应绿色计算科学发展新要求.因此,在理论上,一方面基于对动态频率和电压等系统参数的精细表述及有效量化,建立面向协同异构计算且易于复用的能效感知云调度模型;另一方面,提出并实现适于超计算机混合体系的多学科背景的元启发式多目标全局优化算法.从技术上解决了面向不同环境目标的云调度实施条件界定及其调度指标(时间、能效)实时变化描述等问题.大量仿真实验结果表明:与3个元启发式云调度器相比,该方法在能效及可扩展等方面优势明显;对于高维实例,整体性能改善分别达到8%,12%和14%.
[Abstract]:Heterogeneous scheduling can make large-scale computing systems aggregate all kinds of resources distributed in wide area in parallel to improve performance. The traditional scheduling goal pursues high performance while neglecting high efficiency, which is far from meeting the new requirements of the development of green computing science. Therefore, on the one hand, based on the fine representation and effective quantification of dynamic frequency and voltage and other system parameters, an energy efficiency perceptual cloud scheduling model for collaborative heterogeneous computing and easy reuse is established; on the other hand, a meta-heuristic multi-objective global optimization algorithm suitable for the multi-disciplinary background of supercomputer hybrid system is proposed and implemented. The problems of defining the implementation conditions of cloud scheduling and describing the real-time change of scheduling index (time, energy efficiency) for different environmental objectives are solved technically. A large number of simulation results show that compared with the three meta-heuristic cloud schedulers, this method has obvious advantages in energy efficiency and scalability, and for high-dimensional examples, the overall performance improvement is 8%, 12% and 14%, respectively.
【作者单位】：鲁东大学信息与电气工程学院;山东大学计算机科学与技术学院;山东省高性能计算中心;山东师范大学信息科学与工程学院;
【基金】：国家自然科学基金(61070017,61272094) 国家高技术研究发展计划(863)(2006AA01A113,2012AA01A306)~~
【分类号】：TP301.6

【相似文献】