网格环境下基于复制的能耗有效依赖任务调度研究

Duplication Based Energy-Efficient Scheduling for Dependent Tasks in Grid Environment

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.

As efficient energy management emerges as an important issue for reliable and green computing, energy-aware scheduling approach is regarded as a promising way since it is practical and low-cost. At present, there exist large challenges in the area of energy-aware scheduling for dependent tasks in grid computing system, because the precedence constraints of applications, massive data transmission, system heterogeneity and the conflict of multiple scheduling indicators should be balanced. In this paper, taking into account all the above factors, we propose ESGDT （energy- efficient scheduling of grid dependent task） algorithm, which aims to reduce energy consumption while optimizing execution time for applications. ESGDT algorithm reduces data transfer time and communication energy consumption through task duplication and progressive ratio metric, and considers complex data dependent relationship between tasks. It also considers the static power of processing element through dynamic power management technique following the trends of chip miniaturization and multi-core technology. Moreover, the condition of task duplication, the computation method of progressive ratio metric, and the rule of task adjustment all properly consider two conflicting scheduling indicators time and energy. ESGDT algorithm also focuses on dynamic and adaptive scheduling issues in total heterogeneous system. Simulation experiments demonstrate that ESGDT algorithm could reduce more energy consumption while not influencing scheduling performance than HEFT, EETDS and HEADUS algorithms.