Patent · US Active

Reinforced computer learning system and method for minimizing power consumed by underutilized data center hardware components

US11996988B1 · kind B1 · utility

1Cited by
6References
20Claims
0Family size

Assignee

Inventors

Key dates

Filing dateApr 27, 2023
Grant dateMay 28, 2024
Priority date
Expiry dateApr 27, 2043

Classification

  • Technology area (CPC H)Electricity
  • CPC primaryH04L67/1012
  • WIPO fieldDigital communication
  • WIPO sectorElectrical engineering

Abstract

An information handling system executing a reinforcement learning (RL) based data center power consumption minimizing system may comprise a network interface device to receive operational telemetry measurements for a plurality of data center processors forming a processing node, including performance and utilization analytics, and a user-specified low-utility threshold value, a hardware processor to predict a future time window in which a utilization rate for the processing node executing a future workload falls below the user-specified low-utility threshold value, the hardware processor to identify an optimal load-balancing instruction for redistributing the future workload across a sub-portion of the processing system associated with a highest reward value for satisfying quality of service (QoS) requirements, and the network interface device to transmit the optimal load-balancing instruction and an instruction to throttle power supplied to a remainder of the processing system.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.