负载聚合对数据中心温度及功耗的影响研究

如今日益增长的数据中心功耗特别是冷却系统功耗已经被日益重视,降低系统功耗能够减少数据中心碳排放。传统负载聚合方法在降低计算功耗的同时,尽量保留数据中心的计算能力,却往往忽略了对冷却功耗的关注。不恰当的负载聚合会带来数据中心峰值温度的升高和冷却功耗的增加。文中首先根据数据中心温度功耗模型探讨了负载聚合的必要条件,接着提出了离线的遗传算法。最后文中以实际在线书店的访问数据进行因特网数据中心仿真,定量分析了基于遗传算法的负载聚合方法对系统温度、性能和功耗的影响。

The Quantitative Research on Temperature and Power Effect of Workload Consolidation to Internet Data Centers

Today the ever-growing power cost especially cooling cost of data centers has been emphasized much for carbon emissions reduction. Traditional Workload consolidation method tried to minimize the computing capacity lost when reducing the computing power, while with little consideration about cooling cost. In this paper we show that workload consolidation could bring hot spot appearance and cooling cost incensement. The power temperature model of data center could be constructed based on heat recirculation matrix which could be got through thermal dynamic simulation. In this paper we first explore the necessary condition for workload consolidation; then the genetic algorithm is proposed to optimize the total energy cost. Finally an online bookstore working trace is applied for Internet data centers simulation. With this test-bed a workload consolidation method with genetic algorithm was evaluated for its effects on system peak inlet temperature, performance and energy cost.