Cloud-Based ML Model Training for Storage Performance Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training machine learning models on production server storage systems is resource and time intensive, especially without GPUs, and often involves redundant efforts since the same models are used across systems.
Innovation Solution
Generating a plurality of trained machine learning models using a cloud computing system to forecast storage performance for storage objects, and selecting the best model for deployment to a target storage system using a champion-challenger process or static ruleset, with updates based on model inference results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained on production server storage systems, then the models can forecast storage performance, but the training process is resource and time intensive
Solution Approach 1:
The patent applies preliminary action by training machine learning models in advance on cloud computing systems before deployment to production storage systems. The models are pre-trained using historical storage performance data and then deployed for inference, eliminating the need for time-consuming training on production systems. This resolves the contradiction by performing the training action beforehand in a suitable environment (cloud) rather than during production operation.
2Measurement precision
If machine learning models are trained on production server storage systems, then the models can forecast storage performance, but the training process consumes significant computational resources
Solution Approach 1:
The patent uses cloud computing systems as an intermediary between data preparation and model deployment. The cloud computing system serves as a mediator that performs the computationally intensive model training task, allowing production storage systems to avoid consuming their own computational resources for training while still benefiting from accurate forecasting models. This resolves the contradiction by transferring the resource-intensive training operation to an external intermediary system.
3Adaptability or versatility
If machine learning models are trained on multiple production storage systems, then the models can be customized for each system, but the training process becomes redundant and wasteful
Solution Approach 1:
The patent applies copying by training a single machine learning model on aggregated data from multiple storage systems in the cloud, then deploying copies of this model to each production storage system. Instead of training separate models on each system (which would be redundant), the approach creates and distributes model copies that are customized through the training process on combined data. This resolves the contradiction by using copying to achieve both customization and efficiency.
4Ease of manufacture
If machine learning models are trained without GPUs, then the training can be performed on typical storage systems, but the training time increases significantly
Solution Approach 1:
The patent uses cloud computing systems as an intermediary that provides GPU-accelerated training capabilities. Production storage systems without GPUs can still achieve fast model training by leveraging the cloud intermediary's computational resources. The cloud system acts as a mediator that bridges the gap between the limited capabilities of typical storage systems and the high-performance training required for efficient model development.
Data Source
AI summary
A method, computer program product, and computing system for generating a plurality of trained machine learning models using a cloud computing system by training a plurality of machine learning models to forecast storage performance for one or more storage objects of a storage system, wherein the cloud computing system is separate from any storage system. A trained machine learning model is selected from the plurality of trained machine learning models to deploy to a target storage system. The trained machine learning model is deployed on the target storage system.