AI-Based Storage Indexing for Faster Input-Output Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing storage systems require multiple media accesses and processing cycles due to the need to traverse traditional search trees for data retrieval, limiting the efficiency of input-output operations.
Innovation Solution
Implementing specially trained artificial intelligence models, particularly small neural networks, to learn the location of data within a volume, thereby eliminating the need to index data using physical addresses and reducing media access requests.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional search trees are used to index data on secondary storage, then data can be organized and accessed, but multiple media access or memory access is required which restricts processing cycles for other operations
Solution Approach 1:
The patent replaces the mechanical traversal of traditional search trees with an artificial intelligence model that directly maps data keys to physical addresses. This substitution eliminates the need for sequential node traversal, reducing multiple media access operations to a single AI model inference, thereby significantly improving input-output operation efficiency while maintaining data access capability
Solution Approach 2:
The AI model is pre-trained offline to learn the mapping between data keys and physical storage addresses. This preliminary training action allows the model to perform rapid lookups during runtime without requiring traversal of the storage structure, thus resolving the contradiction between maintaining data access organization and improving operation efficiency
2Measurement precision
If traditional search trees traverse to locate data blocks, then data can be found, but a number of input-output operations are required which limits processing cycles
Solution Approach 1:
The patent substitutes the time-consuming mechanical traversal process with an AI model that performs direct key-to-address mapping. The model maintains measurement precision by accurately locating data blocks while reducing processing cycle consumption from multiple sequential access operations to a single neural network inference operation
Solution Approach 2:
The AI model creates a learned copy of the storage structure's access patterns during offline training. This copied knowledge allows the model to rapidly determine data locations without physically traversing the storage hierarchy, thus maintaining location accuracy while minimizing time loss
Data Source
AI summary
The present technology relates to identifying one of a plurality of artificial intelligence models based on a received key value. The received key value is applied to the identified one of the artificial intelligence models which is trained to output one of a plurality of actual index values. The one of the actual index values is compared to a range within the actual index values to determine when the one of the actual index value points to the data block. A received input-output operation is performed on the data block when the one of the actual index values is determined to point to the data block.