Storage Access Request Processing with Migrated AI Models

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Model retraining on new storage devices consumes significant time, affecting the efficiency of cache prefetch, load identification, and intelligent scheduling functions.

Innovation Solution

Migrate the AI model trained by the old storage device to the new storage device, allowing the new device to reuse the existing model, thereby avoiding the need for retraining and accelerating the provision of cache prefetch, load identification, and intelligent scheduling functions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the new storage device performs model retraining based on historical IO requests, then the AI model can be obtained for cache prefetch, load identification, and intelligent scheduling functions, but the model retraining consumes significant time, affecting the efficiency of providing these functions

Engineering Contradiction:
ImproveAI model accuracyVSAvoidmodel retraining time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The AI model is trained in advance on the old storage device before data migration occurs. The pre-trained model parameters are then migrated together with the data to the new storage device, eliminating the need for retraining and enabling immediate use of cache prefetch, load identification, and intelligent scheduling functions.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The AI model parameters trained on the old storage device are copied to the new storage device. Instead of retraining the model from scratch on the new device, the existing model parameters are transferred and reused, significantly reducing the time required to obtain functional AI capabilities.

Inventive Principle:
Principle #26Copying

2Productivity

If the new storage device performs model retraining, then the AI model can provide cache prefetch, load identification, and intelligent scheduling functions, but the time overhead reduces productivity

Engineering Contradiction:
Improvefunction provision efficiencyVSAvoidmodel retraining time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The AI model training is performed in advance on the old storage device before migration. The pre-trained model is transferred to the new storage device, allowing the new device to immediately provide cache prefetch, load identification, and intelligent scheduling functions without undergoing time-consuming retraining, thus improving productivity from the start.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The trained AI model parameters are copied from the old storage device to the new storage device during data migration. This copying approach enables the new device to inherit the AI capabilities of the old device, eliminating retraining time overhead and immediately achieving high productivity in providing storage optimization functions.

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3995971B1Access request processing method, apparatus, device, and storage medium
Publication Date: 2025.10.29 HUAWEI TECH CO LTD
  • EP3995971B1 patent drawingFigure 1~2
  • EP3995971B1 patent drawingFigure 3
  • EP3995971B1 patent drawingFigure 4

AI summary

This application provides an access request processing method and apparatus, a device, and a storage medium, and pertains to the field of storage technologies. In embodiments of this application, in a data migration scenario, data of an old storage device and an AI model obtained by the old storage device through training are both migrated from the old storage device to a new storage device. If the new storage device receives an access request, a prediction result of the AI model for the access request can be obtained by reusing the AI model trained by the old storage device. This avoids time overheads caused by model retraining by the new storage device, and improves efficiency of the new storage device in providing a function by using the AI model.