Abstract
When a processor - central processing unit (CPU) - in a data center hits a power limit, the processor automatically throttles the operating frequency. Since such CPUs are operating at their power limits, no fault mechanisms are tripped, such that the only consequence of such slow CPUs is degraded user experience. This disclosure describes techniques to screen a fleet of CPUs to identify and service CPUs that are at their power-consumption limits and are throttling their operating frequencies. During CPU throttling, the operating frequency drops but the current consumption remains steady, such that the ratio of current consumption to CPU frequency (Amps-to-MHz ratio) increases. A throttling CPU can be detected by setting a threshold on the Amps-to-MHz ratio of the CPU. A CPU identified as slowing down can be swapped for another CPU, power cycled, reset, subjected to a root-cause analysis, etc.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
n/a, "Numerical Method for Identification of Processors with High Power Consumption and Frequency Throttling", Technical Disclosure Commons, (March 28, 2025)
https://www.tdcommons.org/dpubs_series/7945