摘要
随着数据中心规模的不断扩大,其能耗巨大的问题也日益突出。分布式数据中心既可以通过计算任务在多个数据中心间的分配实现功率的转移,也可以通过单个数据中心的功率控制实现功耗和计算时延的均衡。这2种优化手段相互耦合,且面临着来自于信息层和能量层的多元不确定性的影响,需要快速可靠的控制手段实现数据中心信息层和能量层的协同优化。该文首先构建了分布式数据中心协同优化调节架构,并分析了多数据中心计算任务分配与单数据中心功率优化的动态特性。其次,构建了基于动态微分方程的信息层和能量层耦合优化问题的统一调节模型。最后,综合考虑系统运营成本及计算时延构建目标函数,引入最优控制理论对该问题求解,实现数据中心信息能量的秒级协同优化控制。仿真结果表明,相比分钟级的控制,基于该策略的快速控制能够较好的追踪可再生能源出力以及计算任务的波动,从而有效提升系统的经济效益及可再生能源就地消纳率。
With the continuous expansion of data centers,the problem of large energy consumption has become increasingly prominent.Distributed data centers can enable power transfer through the distribution of computing tasks among multiple data centers and realize the balance between power consumption and computing delay through the power control of a single data center.Scheduling of computing tasks and power control of data center interact with each other,and their control effects are affected by multiple uncertainties.Therefore,a fast and reliable control method is required for realizing the collaborative optimization of the information and energy layers of the data center.First,a distributed data center collaborative optimization architecture is constructed.Then,the dynamic characteristics of multiple data center computing task allocation and single data center power optimization are analyzed based on the dynamic differential equation,and a unified adjustment model of the coupling optimization problem is constructed.Given the system operating cost and computing delay in constructing the objective function,the optimal control theory is introduced to solve the problem and realize the second-level collaborative optimal control of the information energy of the data center.Simulation results show that the high-frequency control based on the proposed algorithm can better track the fluctuation of renewable energy output and calculation tasks than the minute-level control and effectively improve the economic benefits of the system and the local consumption rate of renewable energy.
作者
刘迪
曹军威
刘明爽
LIU Di;CAO Junwei;LIU Mingshuang(Department of Automation,Tsinghua University.Bejjing 100084,China;Beijing National Research Center for Information Science and Technology,Tsinghua University.Beijing 100084,China;Shenzhen Tencent Computer System Co.,Ltd,,Shenzhen 518057,China)
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2022年第12期1864-1874,共11页
Journal of Tsinghua University(Science and Technology)
基金
腾讯基础平台技术犀牛鸟专项研究计划(T102-TEG-2021110400001)。
关键词
分布式数据中心
微分方程
协同优化
可再生能源
最优控制
distributed data center
differential equation
collaborative optimization
renewable energy
optimal control
