Utilizing Conducted Electromagnetic Interference (EMI) for Low-Cost Server-Level Power Monitoring in Data Centers

对互联网和云计算的日益依赖使得数据中心在现代社会中不可或缺。然而，运营这些数据中心需要大量的电力，这导致了全球能源危机。因此，管理数据中心的用电量以实现能效对于可持续的网络基础设施至关重要。为此，服务器电源监控发挥着至关重要的作用，因为它为挖掘数据中心运营的优化潜力提供了有价值的见解。此外，精细的电源监控对于保护数据中心免受因过载和识别异常服务器行为或恶意网络攻击而导致的停机和停机至关重要。然而，由于每台服务器上的专用电能表成本较高，服务器级电源监控尚未被广泛采用。该项目通过开发一种新型的低成本的服务器级电源监控系统来打破成本障碍，该系统不需要在每台服务器上都有专用的电源传感器。相反，它从数据中心配电系统的单点电压测量中提取多台服务器的功耗数据。除了节省成本外，通过该项目实现的细粒度电力监控还可以激励数据中心运营商采用更高效、更环保的技术，从而产生深远的影响。该项目的研究成果和工具将转化为本科生和研究生课程的教材。该项目还将通过实验室访问和暑期实习，纳入涉及妇女、STEM中代表性不足的少数群体以及残疾人的研究活动。该项目的新型功率计量方法的亮点在于，服务器电源中的功率因数校正(PFC)电路通过高频(千赫级)传导的电磁干扰(EMI)向数据中心电源网络显示其功率使用信息。从服务器进行的EMIS可以从数据中心电压测量，以进行服务器级电源监控。然而，在数据中心实现传导式EMI监测系统存在几个技术挑战。数据中心环境中传导EMI的产生和传播尚未被很好地理解，数据中心架构和环境如何影响服务器EMI也未被探讨。此外，从单个传感器提取多个服务器的功耗数据带来了源分离的根本挑战。最后，高频电磁干扰的提取和源分离需要大量和高精度的计算，而开发实时和低成本的传感系统并不是一件容易的事情。鉴于这些挑战，该项目的研究工作是通过三个研究推进来组织的。推力1的目标是通过对传导EMI的产生、传播和提取进行基础性研究，弥合数据中心传导EMI方面的知识差距。推力2专注于开发新的源分离算法，以从单个传感器中提取服务器级的功率。推力3开发了专为实时监测量身定做的计算和成本效益高的传感器硬件和软件。该项目为源分离技术和高效实时传感器数据处理的知识库做出了贡献。该项目推进了对数据中心电力网络传导EMI的基础性认识。此外，传导EMI是由需要调节功率的许多不同类型的设备产生的。因此，该项目的成果可以促进在其他领域处理传导电磁干扰问题的新技术的发展。为了促进这样的发展，这个项目的所有实验数据、代码和设计都将公之于众。这个奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The increasing dependence on the Internet and cloud computing has made data centers indispensable in modern society. However, operating these data centers requires a massive amount of electricity that contributes to the global energy crisis. Therefore, managing data center power usage to achieve energy efficiency is critical for a sustainable cyberinfrastructure. Towards that, server power monitoring plays a crucial role by offering valuable insights for exploring the optimization potential in data center operation. Moreover, fine-grained power monitoring can be vital in safeguarding data centers from outages and downtimes due to overloading and identifying anomalous server behaviors or malicious cyberattacks. However, server-level power monitoring has not been widely adopted due to the high cost associated with dedicated power meters at every server. This project breaks the cost barrier by developing a novel low-cost server-level power monitoring system that does not need dedicated power sensors at every server. Instead, it extracts multiple servers' power consumption data from single-point voltage measurements of the data center power distribution system. Beyond cost savings, the fine granular power monitoring achieved through this project can have a far-reaching impact by motivating data center operators to adopt more efficient and environment-friendly techniques. This project's research results and tools will be translated into educational materials for undergraduate and graduate courses. The project will also incorporate research activities involving women, underrepresented minority groups in STEM, and persons with disabilities through lab visits and summer internships. The enabling insight for this project's novel power metering approach is that the power factor correction (PFC) circuits in the server power supply reveal their power usage information into the data center power network through high-frequency (kHz level) conducted electromagnetic interference (EMI). The conducted EMIs from servers can be measured from the data center voltage for server-level power monitoring. However, there are several technical challenges to realizing a conducted EMI monitoring system in data centers. The generation and propagation of conducted EMI in a data center environment are not well understood and how the data center architecture and the environment affect the server EMI is unexplored. In addition, extracting many servers' power consumption data from a single sensor poses fundamental challenges of source separation. Lastly, high-frequency EMI extraction and source separation require extensive and high-precision computation, while developing a real-time and cost-effective sensing system is non-trivial. In light of these challenges, this project's research efforts are organized through three research thrusts. The goal of Thrust 1 is to bridge the knowledge gap in conducted EMI in data centers through foundational research into conducted EMI generation, propagation, and extraction. Thrust 2 focuses on developing novel source separation algorithms to extract server-level power from a single sensor. Thrust 3 develops computation and cost-effective sensor hardware and software tailored to real-time monitoring. This project contributes to the knowledge base of source separation techniques and efficient real-time sensor data processing. The project advances the foundational understanding of conducted EMI in data center power networks. Moreover, conducted EMI is generated by many different types of equipment requiring regulated power. Hence, this project's outcome can stimulate the development of novel techniques dealing with conducted EMI issues in other domains. To foster such developments, all experiment data, codes, and designs of this project will be made publicly available.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.