ML-Based Data Storage Prediction for Secondary Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional information management systems rely on static storage policies that fail to adapt to data usage patterns, leading to inefficient data storage and retrieval, resulting in reduced performance and increased memory usage, as they struggle to determine which data to store or recall from secondary storage devices effectively.
Innovation Solution
An information management system utilizing machine learning to predict data storage and retrieval decisions by training data storage and recall models based on monitored usage data and context information, allowing for dynamic determination of what data to store or recall from secondary storage devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If static storage policies are used to determine data storage and retrieval, then system simplicity is maintained, but data storage efficiency and operational speed deteriorate
Solution Approach 1:
The patent applies dynamics by transitioning from static storage policies to dynamic machine learning models that adapt to changing data usage patterns. The system continuously monitors data access behavior and retrains models to optimize storage decisions in real-time, making the storage system responsive to actual usage rather than relying on predetermined rules.
Solution Approach 2:
The system implements feedback mechanisms by monitoring data usage patterns and using this information to retrain machine learning models. The monitored usage data feeds back into the model training process, allowing the system to learn from actual behavior and improve storage decisions iteratively, creating a closed-loop optimization system.
2Productivity
If machine learning models are used to predict data storage and retrieval, then storage efficiency and operational speed improve, but system complexity and computational resource usage increase
Solution Approach 1:
The system applies self-service by enabling the machine learning models to automatically make storage and retrieval decisions without manual intervention. The models autonomously predict which data should be stored or recalled based on learned patterns, and the system self-optimizes through automated retraining using monitored usage data, reducing the need for complex manual management.
Solution Approach 2:
The system performs preliminary action by using machine learning models to predict future storage and retrieval needs before actual requests occur. The models analyze historical patterns and proactively determine which data should be pre-positioned in storage, allowing the system to prepare in advance rather than reacting to requests in real-time.
3Loss of substance
If machine learning models are used to predict data storage, then unnecessary data storage is reduced, but memory usage during model training and operation increases
Solution Approach 1:
The system applies parameter changes by adjusting model architecture and training parameters to optimize the balance between prediction accuracy and memory consumption. The patent mentions configuring model parameters such as learning rates, batch sizes, and architecture details to achieve efficient operation with reduced memory footprint while maintaining effective storage predictions.
Data Source
AI summary
An information management system is provided herein that uses machine learning (ML) to predict what data to store in a secondary storage device and/or when to perform the storage. For example, a client computing device can be initially configured to store data in a secondary storage device according to one or more storage policies. A media agent in the information management system can monitor data usage on the client computing device, using the data usage data to train a data storage ML model. The data storage ML model may be trained such that the model predicts what data to store in a secondary storage device and/or when to perform the storage. The client computing device can then be configured to use the trained data storage ML model in place of the storage polic(ies) to determine which data to store in a secondary storage device and/or when to perform the storage.