Cluster Scheduler Predictive Load Balancing

Solution Overview

Problem

Current resource load balancing control methods in cloud centers are ineffective, leading to invalid migrations due to reliance on historical data and irregular load patterns, which result in incorrect decision-making and imbalanced resource allocation.

Innovation Solution

A method that predicts performance data for each cluster node over a preset time period, calculates a standard deviation to assess resource load balance, and determines migration solutions based on these predictions to balance resource loads, using machine learning models trained with historical data to improve migration validity and accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If resource load balancing control is performed based on historical data usage, then resource allocation can be optimized, but invalid migrations occur when load patterns are irregular or temporary

Engineering Contradiction:

Improvemigration validityVSAvoidload balance assessment accuracy

Core Design Contradiction:

ReliabilityVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by predicting future performance data before making migration decisions. The system predicts performance data for a preset time period ahead of time, allowing the cluster scheduler to assess whether a migration is truly necessary before executing it. This prevents invalid migrations caused by temporary load spikes or irregular patterns by looking ahead rather than reacting to historical data alone.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback by using predicted performance data to continuously monitor and adjust resource allocation. The cluster scheduler compares predicted performance data against thresholds and migration criteria, creating a closed-loop system that can correct previous incorrect decisions. This feedback mechanism allows the system to learn from past migration outcomes and improve future decision-making accuracy.

Inventive Principle:

Principle #23Feedback

2Productivity

If virtual machine migration is performed to balance resource load, then resource distribution improves, but system stability decreases due to frequent or incorrect migrations

Engineering Contradiction:

Improveresource utilization efficiencyVSAvoidcluster system stability

Core Design Contradiction:

ProductivityVSStability of the object's composition

Solution Approach 1:

The patent applies preliminary action by predicting future performance data before making migration decisions. The system predicts performance data for a preset time period ahead of time, allowing the cluster scheduler to assess whether a migration is truly necessary before executing it. This prevents invalid migrations caused by temporary load spikes or irregular patterns by looking ahead rather than reacting to historical data alone.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback by using predicted performance data to continuously monitor and adjust resource allocation. The cluster scheduler compares predicted performance data against thresholds and migration criteria, creating a closed-loop system that can correct previous incorrect decisions. This feedback mechanism allows the system to learn from past migration outcomes and improve future decision-making accuracy.

Inventive Principle:

Principle #23Feedback

3Measurement precision

If prediction models are trained using historical data, then migration accuracy improves, but data processing complexity increases

Engineering Contradiction:

Improveperformance prediction accuracyVSAvoiddata processing complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by transforming historical data into predicted performance data through trained models. The system changes the state of data from raw historical records to processed predictions, making the data more suitable for migration decision-making. This transformation simplifies the decision-making process while maintaining high prediction accuracy through the trained models.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS10924535B2Resource load balancing control method and cluster scheduler

Publication Date: 2021.02.16 HUAWEI TECH CO LTD

AI summary

Embodiments of the present invention disclose a resource load balancing control method and a cluster scheduler. The method includes: predicting performance data that is of an application deployed on each cluster node and that is in a preset time period; calculating a first standard deviation of the cluster system according to the predicted performance data of each cluster node; when the first standard deviation of the cluster system is greater than a preset threshold, determining an application migration solution according to a resource load balancing rule; and sending a application migration solution to the cluster application manager, to trigger the cluster application manager to perform resource load balancing control on the cluster system according to a application migration solution.