Heterogeneous Power Management for Data Center Servers

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional power management systems in data centers assume homogeneous server power dynamics, leading to sub-optimal processing performance due to the heterogeneity in power consumption among servers, which is influenced by factors like process variations, age, and cooling efficacy.

Innovation Solution

A heterogeneous-aware power management system evaluates each computing resource's idle and peak power consumption metrics to dynamically configure power usage, implementing server consolidation, workload allocation, and power capping based on individual power dynamics to efficiently meet power budget constraints while preserving processing capacity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional homogeneous power management approaches are used, then power budget control is achieved, but processing performance is sub-optimal due to ignoring server power dynamics heterogeneity

Engineering Contradiction:

Improveprocessing performanceVSAvoidpower efficiency

Core Design Contradiction:

ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent segments the data center into individual server units, each with its own measured power dynamics characteristics. Instead of treating all servers as homogeneous, the system divides them into distinct entities with unique power profiles, allowing differentiated power management strategies to be applied to each server based on its specific characteristics.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by assigning different power management parameters and strategies to different servers based on their individual power dynamics. Each server receives customized power allocation, consolidation, and capping decisions tailored to its specific idle and peak power consumption characteristics rather than applying uniform management across all servers.

Inventive Principle:

Principle #3Local quality

2Quantity of substance

If server consolidation is implemented without considering individual power dynamics, then power budget is reduced, but processing capacity is lost due to sub-optimal consolidation decisions

Engineering Contradiction:

Improvepower budget utilizationVSAvoidprocessing capacity

Core Design Contradiction:

Quantity of substanceVSProductivity

Solution Approach 1:

The patent implements feedback by continuously measuring and monitoring the actual power consumption of each server, then using this measured data to inform consolidation decisions. The system feeds back the power dynamics information to the power management controller, which adjusts consolidation strategies based on real or near-real-time power consumption patterns rather than relying on theoretical or average values.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The patent applies preliminary action by pre-characterizing each server's power dynamics through measurement before making consolidation decisions. The system performs power consumption measurements and establishes power profiles in advance, so that when consolidation decisions are made, they are based on pre-acquired knowledge of each server's power characteristics rather than making decisions without this information.

Inventive Principle:

Principle #10Preliminary action

3Use of energy by stationary object

If power capping is applied uniformly across all servers, then power consumption is controlled, but processing performance deteriorates due to heterogeneity in power dynamics

Engineering Contradiction:

Improvepower consumption controlVSAvoidprocessing performance

Core Design Contradiction:

Use of energy by stationary objectVSProductivity

Solution Approach 1:

The patent applies parameter changes by adjusting power cap levels individually for each server based on its specific power dynamics characteristics. Instead of applying a uniform power cap to all servers, the system modifies the power cap parameters for each server according to its measured idle and peak power consumption patterns, allowing optimal performance within power constraints.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS10168762B2Power management for heterogeneous computing systems

Publication Date: 2019.01.01 ADVANCED MICRO DEVICES INC

AI summary

A computing system includes a set of computing resources and a datastore to store information representing a corresponding idle power consumption metric and a corresponding peak power consumption metric for each computing resource of the set. The computing system further includes a controller coupled to the set of computing resources and the datastore. The controller is to configure the set of computing resources to meet a power budget constraint for the set based on the corresponding idle power consumption metric and the corresponding peak power consumption metric for each computing resource of the set.