Compute Node Power Management via Component Utilization Analysis

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Datacenters face high power consumption and associated expenses due to compute nodes, with existing power management methods often reducing performance by imposing caps on components, which can lead to inefficient energy use and thermal issues.

Innovation Solution

A method that collects component utilization data to identify underutilized components during previous workload executions and throttles them to prevent exceeding a threshold utilization level during subsequent executions, optimizing power usage without compromising performance.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If power caps are imposed on compute nodes or components, then power consumption is reduced, but workload performance deteriorates

Engineering Contradiction:

Improvepower consumptionVSAvoidworkload performance

Core Design Contradiction:

Use of energy by moving objectVSProductivity

Solution Approach 1:

The system performs preliminary analysis of component utilization data from previous workload executions to identify which components are underutilized before imposing power caps. This preliminary identification ensures that only non-critical components are throttled, preventing performance degradation while achieving power reduction goals

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system continuously monitors component utilization data and uses this feedback to dynamically adjust power management decisions. By analyzing utilization patterns from previous executions and real-time data, the system adapts its throttling strategy to maintain performance while reducing power consumption

Inventive Principle:

Principle #23Feedback

2Productivity

If component utilization is increased to improve performance, then productivity increases, but power consumption and heat generation increase

Engineering Contradiction:

Improveworkload performanceVSAvoidpower consumption

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The system applies different power management strategies to different components based on their individual utilization characteristics. Instead of uniformly throttling all components, it selectively identifies underutilized components and applies power caps only to those, while allowing heavily utilized components to operate at full capacity

Inventive Principle:

Principle #3Local quality

3Use of energy by moving object

If uniform power caps are applied to all components, then power consumption is reduced, but system complexity increases and performance optimization decreases

Engineering Contradiction:

Improvepower consumptionVSAvoidmanagement complexity

Core Design Contradiction:

Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The system segments the compute node into individual components and analyzes each component's utilization independently. This segmentation allows for targeted power management decisions on a per-component basis, simplifying the overall management complexity while achieving better power optimization than uniform caps

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS10819607B2Reducing power consumption of a compute node experiencing a bottleneck

Publication Date: 2020.10.27 LENOVO GLOBAL TECHNOLOGIES SWITZERLAND INTERNATIONAL GMBH

AI summary

A method includes obtaining component utilization data for multiple components of a compute node during at least one previous execution of a workload. The method further includes using the component utilization data to identify a first component having a utilization level that is less than a threshold utilization level during the at least one previous execution of the workload, wherein the first component is one of the multiple components of the compute node. The method still further includes, during a subsequent execution of the workload on the compute node, throttling the first component to prevent the first component from exceeding the threshold utilization level.