算力-电力-热力协同视角下数据中心高密机柜全栈联动研究

作者：Yongzhen Wang, Kuo Song, Kai Han, Hao Luo, Jiayu Lin, Te Han, Yiming Wei, Zixian Li · 发表于：Scientia Sinica Technologica · 年份：2024 · DOI：10.1360/sst-2024-0170 · 被引用次数：1 · 研究领域：Big Data and Digital Economy、Cloud Computing and Resource Management、Parallel Computing and Optimization Techniques

The continuous development of intelligent computing is driving the high density of computing power, leading to rapid growth in power consumption and heat dissipation density of data center (DC) cabinets. Currently, most of the energy-saving control methods for DCs are still limited to the optimization of a single chain of the upper computing equipment or the lower heat dissipation equipment, and the global optimization potential of temperature flexible control on cabinet computing and electric power consumption has not been deeply explored. Therefore, based on global optimization and dynamic scheduling, this paper proposes the concept of full-stack linkage of high-density DCs under the coordination of computing, electric, and thermal powers. Considering as examples the server power consumption and efficiency model, two-phase liquid cooling technology, and dynamic control technology involved in full-stack linkage, the related research status is analyzed, and the model construction and solution method for full-stack linkage of DCs under temperature flexible control are discussed. Accordingly, the coupling influence mechanism of operating temperature on computing, electric, and thermal powers is analyzed, and the research trend of full-stack linkage of DCs based on model-guided feedback correction is discussed.