Neural Network Object Storage Migration Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object storage systems face challenges in efficiently managing data distribution across multiple storage devices due to the time-consuming nature of heuristic algorithm calculations, which hinders quick response to changes and optimal quality of service (QoS).
Innovation Solution
The implementation of a migration control method using multiple neural networks to infer optimal object assignments, where the first neural network infers reference I/O throughputs based on total target I/O throughputs, and the second neural network determines target objects and storage devices for migration based on object points calculated from target and reference I/O throughputs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If heuristic algorithm calculations are used for object assignment, then the system can manage data distribution, but the calculation time becomes too long to respond quickly to changes
Solution Approach 1:
The patent replaces traditional heuristic algorithms with neural network-based inference systems. The first neural network predicts reference I/O throughputs based on storage device characteristics, while the second neural network determines optimal object migration decisions by processing object points and storage state information. This substitution of mechanical calculation systems with neural network inference enables rapid response to system changes while maintaining quality of service optimization.
2Reliability
If multiple neural network inferences are performed for optimal object assignment, then the quality of service improves, but the system complexity increases
Solution Approach 1:
The patent divides the object assignment problem into two separate neural network inference stages. The first neural network handles the prediction of reference I/O throughputs for different storage devices, while the second neural network processes object points and storage state information to determine migration decisions. This segmentation of the complex optimization problem into two specialized inference tasks reduces the complexity of each individual network while achieving optimal quality of service through their coordinated operation.
Data Source
AI summary
An object storage system, a migration control device, and a migration control method are provided. The migration control method includes observing a plurality of total target input/output (I/O) throughputs, inferring a plurality of reference I/O throughputs by using a first neural network, observing a plurality of target I/O throughputs with respect to an i-th storage device, determining a target object to be migrated and a target storage device, in which the target object will be stored, by using a second neural network, and generating a command instructing to migrate the target object to the target storage device.