CRII: CNS: Auction Mechanism Design for Energy-Efficient High Performance Computing

高性能计算（HPC）系统（诸如超级计算机）通常是包含数千个服务器节点的大型基础设施，这些服务器节点可以以快速且高效的方式执行计算。HPC系统在运行期间会消耗大量的功率。例如，目前排名靠前的超级计算机在峰值运行期间可以消耗数十兆瓦的电力。作为功率消耗增加的直接结果，能量成本已经成为HPC系统的操作的总成本的主要组成部分。为了实现HPC的能源可持续性，该项目计划开发新的模型，以降低能源成本并有助于电力系统的稳定性。本项目的主要目标有三个：（1）开发机器学习模型来预测并行应用程序的功率和性能;（2）开发拍卖机制模型，通过HPC用户的集体能源减少来降低HPC系统的能源成本，同时将可再生能源发电纳入模型中;（3）通过仿真实验和验证所提出的拍卖机制模型。总体而言，该项目有望降低大规模系统的能源成本，实现电网的节能和稳定，推动HPC能效水平的发展，平衡HPC的能效平衡。通过这样做，该项目将增加HPC系统在可持续计算中的参与。拟议的研究将使HPC系统与电网系统密切互动，并根据电价变化和可再生能源发电实现基于反馈的节能。该项目将增加研究生和本科生的研究参与。此外，该项目将在并行和高性能计算以及节能计算领域对学生进行培训和教育。此外，通过各种推广活动和研究参与，该项目计划通过涉及代表性不足的群体来促进计算的多样性。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

High performance computing (HPC) systems (such as supercomputers) are generally large infrastructures containing thousands of server nodes that can perform computations in a fast and efficient manner. HPC systems can consume an enormous amount of power during their operation. For example, current top-ranked supercomputers can consume tens of megawatts of power during peak operation. As a direct consequence of power consumption increase, energy cost has become a major component of the overall cost of the operation of an HPC system. To achieve energy sustainability in HPC, this project plans to develop novel models to reduce energy cost and contribute to the power system stability. There are three primary objectives of this project: (1) develop machine learning models to predict the power and performance of parallel applications; (2) develop an auction mechanism model to reduce HPC system’s energy cost via collective energy reduction of HPC users, while incorporating the renewable energy generation into the model; and (3) experiment and validate the proposed auction mechanism model via simulation. Overall, the project is expected to reduce the energy cost of large-scale systems, as well as to achieve power grid energy conservation and stability.This project will contribute towards advancement of the state-of-the-art in energy-efficiency of HPC, as well as to balance the energy-performance trade-offs in HPC. In doing so, this project will increase HPC system’s participation in sustainable computing. The proposed research will enable HPC systems to closely interact with the power grid system, and enable feedback-based energy reduction based on electricity price variation and renewable energy generation. This project will increase research participation of both graduate and undergraduate students. Additionally, the project will train and educate students in the area of parallel and high performance computing, and energy-efficient computing. Furthermore, through various outreach activities and research involvement, the project plans to promote diversity in computing by involving underrepresented groups.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.