网格环境下基于复制的能耗有效依赖任务调度研究

发布时间：2018-11-11 16:37

【摘要】：随着能耗管理成为可靠和绿色计算的重要课题,能耗感知调度方法以其低成本和可行性引发关注.目前,网格环境下依赖任务的能耗感知调度研究具有极大的挑战性,其需要平衡应用的优先约束性、海量数据传输、系统的异构性和不同性能指标的冲突性的关系.提出的网格依赖任务的能耗有效调度(energy-efficient scheduling of grid dependent tasks,ESGDT)算法旨在优化应用执行时间的前提下降低应用执行能耗,能有效解决上述问题.通过任务复制和渐进比例因子减少通信时间和通信能耗,同时兼顾应用复杂的数据依赖关系;适应芯片微型化和多核技术的发展趋势,采用动态电源管理技术减少任务执行的静态能耗;任务复制条件、渐进比例因子和微调原则均适时兼顾时间和能耗两个相互冲突的调度指标,并提出自适应和动态映射方法适应异构计算环境.模拟实验表明,较HEFT,EETDS和HEADUS算法,ESGDT算法不仅没有影响调度的时间性能,还可进一步降低应用执行能耗.
[Abstract]:As energy consumption management has become an important issue in reliable and green computing, energy-aware scheduling has attracted attention due to its low cost and feasibility. At present, the research of task-dependent energy-aware scheduling in grid environment is very challenging. It needs to balance the priority constraint of application, the mass data transmission, the heterogeneity of the system and the conflict of different performance indicators. An energy-efficient scheduling (energy-efficient scheduling of grid dependent tasks,ESGDT) algorithm based on grid dependent tasks is proposed to reduce the energy consumption of application execution on the premise of optimizing application execution time, which can effectively solve the above problems. The communication time and communication energy consumption are reduced by task replication and progressive scaling factor, and the complex data dependency is also taken into account. To adapt to the trend of microchip miniaturization and multi-nuclear technology, dynamic power management technology is used to reduce the static energy consumption of task execution. Task replication conditions, progressive scaling factors and fine-tuning principles all take into account time and energy consumption, and adaptive and dynamic mapping methods are proposed to adapt to heterogeneous computing environments. Simulation results show that compared with HEFT,EETDS and HEADUS algorithms, ESGDT algorithm not only has no effect on scheduling time performance, but also can further reduce application execution energy consumption.
【作者单位】：山东电力集团公司电力科学研究所;山东大学计算机科学与技术学院;
【基金】：国家自然科学基金项目(61070017) 山东大学研究生自主创新基金项目(yzc09063)
【分类号】：TP393.06

【相似文献】