Container Cluster Auto Scaling With ML-Based Configuration Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Determining the optimal configuration for container-based clusters is challenging due to the need to balance cost, resource consumption, and application performance/availability, and conventional methods are often subjective, non-scalable, and require trial-and-error, leading to inefficiencies and increased operational complexity.
Innovation Solution
Employing machine learning models to generate and evaluate candidate configurations for container-based clusters, predicting future performance and resource utilization, and selecting a target configuration that optimizes workload performance and availability while minimizing cost and complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional manual methods are used to determine cluster configuration, then expert knowledge can be applied, but the process is subjective, non-repeatable, and not scalable
Solution Approach 1:
The patent replaces manual expert assessment with an automated machine learning system that objectively evaluates cluster configurations. The ML model processes workload patterns, resource utilization metrics, and performance data to generate optimized configurations, eliminating subjectivity and enabling scalable automation while maintaining or improving configuration accuracy.
2Manufacturing precision
If trial-and-error process is used to determine configuration, then various configurations can be tested, but many pods are terminated/restarted and resources are wasted
Solution Approach 1:
The patent uses machine learning models to predict the outcomes of different cluster configurations before actually deploying them. By simulating and evaluating multiple candidate configurations in advance, the system identifies the optimal configuration without requiring extensive trial-and-error testing, thereby preventing pod terminations, restarts, and resource waste that would occur with conventional approaches.
3Adaptability or versatility
If configurations are constantly updated to match changing application requests, then application performance is maintained, but operational complexity increases
Solution Approach 1:
The patent implements a self-service automated system that continuously monitors workload patterns and resource utilization, then autonomously generates and applies configuration updates without human intervention. The machine learning model detects when changes are needed and automatically adjusts cluster configurations, maintaining application performance while eliminating the operational complexity of manual configuration management.
4Reliability
If more resources are allocated to ensure application availability, then application performance is improved, but cost increases
Solution Approach 1:
The patent dynamically adjusts cluster configuration parameters based on actual workload patterns and performance requirements. The machine learning model analyzes historical data to determine the minimum resources needed to maintain application availability, then optimizes resource allocation by adjusting pod counts, replica sets, and node configurations. This ensures sufficient resources for reliability while preventing over-provisioning and reducing unnecessary resource consumption and cost.
Data Source
AI summary
The disclosure provides a method for determining a target configuration for a container-based cluster. The method generally includes determining, by a virtualization management platform configured to manage components of the cluster, a current state of the cluster, determining, by the virtualization management platform, at least one of performance metrics or resource utilization metrics for the cluster based on the current state of the cluster, processing, with a model configured to generate candidate configurations recommended for the cluster, the current state and at least one of the performance metrics or the resource utilization metrics and thereby generate the candidate configurations, calculating a reward score for each of the candidate configurations, selecting the target configuration as a candidate configuration from the candidate configurations based on the reward score of the target configuration, and adjusting configuration settings for the cluster based on the target configuration to alter the current state of the cluster.