异构计算系统中弹性节能调度策略研究

目前,节能已成为异构计算系统中减少电量开销、提高系统可靠性和保护环境的重要研究内容.传统的节能调度策略侧重于研究如何节能而忽略了用户对任务完成时间的期望,使得任务执行效果受到较大影响.特别是当系统负载较重时,由于电压调节缺乏自适应性,导致在某些情况下(如应急服务)的任务执行效果不可容忍.文中提出一种弹性节能调度策略(Elastic Energy-Aware Scheduling,EEAS),用于动态调度异构计算系统中非周期、独立任务.EEAS策略根据系统负载情况在系统节能与用户期望之间进行权衡,即当系统负载较重时,EEAS优先考虑用户期望,通过动态调整计算节点局部队列中等待任务的执行电压提高任务完成率;当系统负载较轻时,EEAS在尽量满足用户期望的基础上最大限度地降低任务执行电压以实现节能.文中通过大量的模拟实验比较了EEAS、GEA、HVEA和LVEA的性能.实验结果表明,EEAS的调度质量优于其他策略,可有效提高系统弹性.

An Elastic Energy-Aware Scheduling Strategy for Heterogeneous Computing Systems

Energy saving has become a major issue for heterogeneous computing systems to minimize electricity cost,improve system reliability and protect environment.Conventional energy-aware scheduling strategies developed on heterogeneous computing systems concentrated on energy savings regardless of the user expected finish times of tasks while making scheduling decisions.As a result,the user expectations by such strategies will be affected greatly especially when the systems are heavily loaded,which results in inferior system adaptivity or,in some situations,(e.g.,emergency service) it is even not tolerated.In this paper,we developed a novel dynamic scheduling strategy named Elastic Energy-Aware Scheduling(EEAS) for aperiodic,and independent tasks on heterogeneous computing systems with dynamic voltage scaling.The EEAS strategy aims at adaptively adjusting voltages according to the system workload,thereby making trade-offs between energy conservation and user expectation.i.e.,when the system is under heavy workload,to meet user expectations,EEAS not only considers the voltage for a new task,but also takes the voltages to run tasks waiting in local queues into account;in contrast,EEAS degrades voltage levels to reduce energy consumption while holding higher user satisfaction rate in terms of user expected finish time.We conducted extensive experiments to compare our EEAS with three schemes — GEA,HVEA and LVEA.Experimental results show that EEAS significantly improves the scheduling quality of others,and is able to effectively enhance the system elasticity.