Cluster Capacity Level Switching for Power and Availability

Solution Overview

Problem

Balancing power consumption and computational availability in computational clusters is challenging, as powering all machines continuously wastes energy while powering them only when needed introduces delays, and finding an optimal balance between these extremes is unclear.

Innovation Solution

Monitoring electrical power consumption and cluster utilization, predicting demand based on job queue length, priority, and historic data to adjust cluster capacity levels, ensuring stability constraints are not violated, and configuring threshold conditions for switching between capacity levels.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If all machines in the cluster are powered continuously, then computational availability is complete, but power consumption increases and energy is wasted when availability is not used

Engineering Contradiction:

Improvecomputational availabilityVSAvoidpower consumption

Core Design Contradiction:

ReliabilityVSLoss of energy

Solution Approach 1:

The patent implements dynamic capacity adjustment by switching between different cluster capacity levels (e.g., 100%, 75%, 50%, 25%) based on predicted demand. The system transitions from a static all-or-nothing power state to a dynamic multi-level capacity model, allowing flexible adjustment of computational resources matched to actual workload predictions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system performs preliminary demand prediction using job queue length, job priority, and historic consumption data before actually allocating resources. This predictive approach allows the system to proactively adjust capacity levels in advance, avoiding the need to keep all machines running continuously while ensuring computational availability is maintained when needed.

Inventive Principle:

Principle #10Preliminary action

2Loss of energy

If machines are powered up only after demand is made, then power consumption is minimized, but delay between demand and computation performance increases

Engineering Contradiction:

Improvepower consumptionVSAvoidresponse delay

Core Design Contradiction:

Loss of energyVSLoss of time

Solution Approach 1:

The system performs preliminary demand prediction using job queue length, job priority, and historic consumption data before actually allocating resources. This predictive approach allows the system to proactively adjust capacity levels in advance, avoiding the need to keep all machines running continuously while ensuring computational availability is maintained when needed.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system continuously monitors cluster utilization, power consumption, and demand patterns to refine its predictions and adjust capacity levels dynamically. This feedback loop ensures that capacity adjustments are based on actual usage patterns and predictive accuracy improves over time, reducing both energy waste and response delays.

Inventive Principle:

Principle #23Feedback

3Loss of energy

If cluster capacity is frequently adjusted to match demand, then power consumption efficiency improves, but stability of the cluster is compromised

Engineering Contradiction:

Improvepower consumption efficiencyVSAvoidcluster stability

Core Design Contradiction:

Loss of energyVSStability of the object's composition

Solution Approach 1:

The patent implements dynamic capacity adjustment by switching between different cluster capacity levels (e.g., 100%, 75%, 50%, 25%) based on predicted demand. The system transitions from a static all-or-nothing power state to a dynamic multi-level capacity model, allowing flexible adjustment of computational resources matched to actual workload predictions.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system performs preliminary demand prediction using job queue length, job priority, and historic consumption data before actually allocating resources. This predictive approach allows the system to proactively adjust capacity levels in advance, avoiding the need to keep all machines running continuously while ensuring computational availability is maintained when needed.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS8788855B2Cluster computational capacity level switching based on demand prediction and stability constraint and power consumption management

Publication Date: 2014.07.22 MICROSOFT TECHNOLOGY LICENSING LLC

AI summary

Power consumption and computational availability of a cluster are automatically managed by monitoring consumption, predicting demand based on a job queue and/or historic data, and checking certain conditions and constraints. Threshold conditions for switching the cluster's capacity level are based on the current level and the predicted demand. Switching levels is avoided when a stability constraint would be violated. Cluster capacity levels can be uniformly or nonuniformly spaced, and may have decreasing gaps between levels as capacity decreases. Capacity level definitions and level switching conditions and constraints may be defaults and/or be provided by an administrator.