Sudden and short-duration power swings in modern supercomputers can have a challenging impact on the voltage of the adjacent power grids. The coming age of exascale supercomputers is expected to bring platforms that are capable of power fluctuations of up to 20 MW in 15 milliseconds (ms) or less (one cycle at 60 Hz). This paper presents a framework that explains these relationships between rapid power flow changes in high performance computing environments (HPC) and power grid characteristics, such as voltage and frequency fluctuations, and offers insights on system improvements that can mitigate large power disturbance issues. Four case studies of HPC facilities illustrate their unique powergrid setting and identify some of the components that help with the mitigation of large power fluctuations. HPC managers were asked about preparing for larger peak dynamic power demand both on their site and within the broader context of their local electrical grid. The data presented was collected based on questionnaires and interviews with these sites. The responses can provide insight on how to address this growing concern and mitigate the local impact of HPC facilities through grid stiffness or resiliency measures.
Stewart, Grant L; Koenig, Gregory A.; Liu, Jingjing; Clausen, Anders; Klingert, Sonja; Bates, Natalie