Containerized Application Resource Optimization With Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems fail to dynamically and automatically optimize application resources and find bottlenecks affecting performance in containerized environments, relying on slower horizontal auto scaling and lacking the ability to optimize resources in a dynamic and automated manner.
Innovation Solution
Implementing a system that uses reinforcement learning to train a model and construct a look-up table in a pre-production environment, continuously refining it with feedback data from a production environment, and leveraging this table to recommend optimal configurations in real-time using a neural network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional horizontal auto scaling is used to optimize application resources, then resource optimization is achieved, but the process is slow and not dynamic
Solution Approach 1:
The system performs preliminary training of the reinforcement learning model using simulated workloads in a pre-production environment before deploying to production. This pre-training establishes a foundation of optimal configurations that enables rapid real-time optimization without requiring slow horizontal scaling operations during production workloads.
Solution Approach 2:
The patent replaces the mechanical horizontal auto-scaling system with an intelligent reinforcement learning-based optimization system. Instead of physically adding or removing container instances through orchestration, the system uses trained AI models to dynamically optimize resource allocation and configuration parameters in real-time, achieving faster response without the overhead of mechanical scaling operations.
2Ease of operation
If manual resource configuration is used, then control over application resources is maintained, but manual processes and errors increase
Solution Approach 1:
The reinforcement learning model autonomously determines optimal resource configurations and deployment parameters without requiring manual intervention. The system self-learns from simulated workloads and production feedback, automatically adjusting container resource allocations, replication factors, and other configuration parameters while maintaining high reliability through continuous learning and validation.
Solution Approach 2:
The system implements continuous feedback loops where production environment performance data is collected and used to refine the reinforcement learning model. This feedback mechanism ensures that the automated configuration process continuously improves accuracy by learning from actual system performance, reducing errors while maintaining ease of operation.
3Measurement precision
If reinforcement learning model training is performed using simulated workloads, then model accuracy is improved, but training time and computational resources increase
Solution Approach 1:
The system performs comprehensive model training in advance using simulated workloads in a pre-production environment. By completing the resource-intensive training phase before production deployment, the system achieves high model accuracy without incurring training time penalties during production operations. The pre-trained model can then rapidly provide optimization recommendations in real-time.
Solution Approach 2:
The system creates copies of production workload patterns as simulated workloads for training the reinforcement learning model. By replicating realistic workload characteristics in a virtual environment, the model achieves high prediction accuracy for production scenarios without requiring actual production data or incurring production system overhead during training.
Data Source
AI summary
Embodiments compare key metrics data representing real-time workloads performed by a computer set representing one or more containers; determine that the key metrics data does not key metrics criteria; in response to the determining, query, from a pre-trained look up table, an optimal configuration for deploying resources to at least one containerized application of the computer set; determine that the optimal configuration is not found from the pre-trained look up table; train a neural network (NN) model by using samples from the pre-trained look up table as training data; determine the optimal configuration using the trained NN model; and deploy the determined optimal configuration for the computer set.