基于模型预测控制的数据中心节能调度算法

发布时间：2018-07-10 19:13

本文选题：模型预测控制 + 反馈控制　；参考：《软件学报》2017年02期

【摘要】：如今日益增长的数据中心能耗,特别是冷却系统能耗已日益受到重视,降低系统能耗能够减少数据中心碳排放.提出了一种基于模型预测控制(model prediction control,简称MPC)的节能调度策略,该策略可以有效地减小数据中心冷却能耗.该方法采用动态电压频率调节技术来调整计算节点频率,从而减少节点间的热循环;所有节点的峰值温度可被保持在温度阈值下,在任务的执行中稳态误差较小.该方法可以通过动态频率调节来抑制由于负载类型变化造成的模型不确定性带来的内部扰动,分析结果表明,基于模型预测的温控算法系统开销较小,具有良好的可扩展性.基于该算法设计的控制器能够有效地降低输入温度,提高数据中心能耗效率.通过在实际数据中心内运行的模拟网上书店,该方法与安全最小热传递算法和传统反馈温控算法这两种经典方法相比,无论是在正常条件下还是在扰动存在的情况下都能取得较好的温度抑制效果,系统性能如吞吐率也达到最大.在相同的负载条件下,该方法能够获得最小的输入峰值温度和最小的冷却能耗.
[Abstract]:Nowadays, increasing energy consumption of data center, especially cooling system, has been paid more and more attention. Reducing energy consumption of data center can reduce carbon emission of data center. An energy saving scheduling strategy based on model predictive control (model prediction control,) is proposed, which can effectively reduce cooling energy consumption in data center. The dynamic voltage frequency regulation technique is used to adjust the frequency of the node to reduce the thermal cycle between nodes. The peak temperature of all nodes can be kept at the temperature threshold and the steady-state error is small in the execution of the task. This method can restrain the internal disturbance caused by the uncertainty of the model due to the change of the load type through dynamic frequency regulation. The analysis results show that the temperature control algorithm based on the model prediction has less system overhead and good scalability. The controller designed based on this algorithm can effectively reduce the input temperature and improve the energy efficiency of the data center. This method is compared with two classical methods, the secure minimum heat transfer algorithm and the traditional feedback temperature control algorithm, by running the simulated online bookstores in the actual data center. Both under normal conditions and in the presence of disturbances, a better temperature suppression effect can be obtained, and the system performance such as throughput is also maximum. Under the same load condition, the method can obtain minimum input peak temperature and minimum cooling energy consumption.
【作者单位】：武汉大学国际软件学院软件工程系;湖北科技学院计算机科学与技术学院;
【基金】：国家自然科学基金(61003185) 湖北省自然科学基金(201FFB04505)~~
【分类号】：TP308

【相似文献】