Vector Retrieval With Two-Level Storage for Lower Shared-Storage Delay

Solution Overview

Problem

Existing vector retrieval solutions face high delays and high throughput issues when applied in shared storage environments, leading to poor adaptation and user experience.

Innovation Solution

A vector retrieval method and apparatus that stores a graph index based on original and compression vectors in a first-level storage resource, where data access performance is better than a second-level storage resource, allowing for decoupled nearest neighbor search processes, reducing synchronous data access operations on the second-level storage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If existing vector retrieval solutions are applied in shared storage environments, then data availability and accessibility are improved, but retrieval delay increases and throughput decreases

Engineering Contradiction:

Improveadaptability to shared storageVSAvoidretrieval delay

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent segments the storage system into two levels: first-level storage for hot data (compression vectors and graph index) and second-level storage for cold data (original vectors). This segmentation allows frequent access operations to occur at the faster first-level storage while less frequently accessed original vectors reside in the shared storage, thereby reducing retrieval delay while maintaining adaptability to shared storage environments.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a compression vector as an intermediary between the query and the original vector in shared storage. The compression vector serves as a proxy that can be quickly accessed and compared, allowing the system to filter candidates without repeatedly accessing the slow shared storage, thus reducing retrieval delay while maintaining compatibility with shared storage architecture.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Adaptability or versatility

If existing vector retrieval solutions are applied in shared storage environments, then data accessibility is improved, but throughput decreases

Engineering Contradiction:

Improveadaptability to shared storageVSAvoidthroughput

Core Design Contradiction:

Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent segments the storage system into two levels: first-level storage for hot data (compression vectors and graph index) and second-level storage for cold data (original vectors). This segmentation allows frequent access operations to occur at the faster first-level storage while less frequently accessed original vectors reside in the shared storage, thereby reducing retrieval delay while maintaining adaptability to shared storage environments.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces a compression vector as an intermediary between the query and the original vector in shared storage. The compression vector serves as a proxy that can be quickly accessed and compared, allowing the system to filter candidates without repeatedly accessing the slow shared storage, thus reducing retrieval delay while maintaining compatibility with shared storage architecture.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Loss of time

If compression vectors are stored in first-level storage and original vectors in second-level storage, then retrieval delay is reduced, but storage complexity increases

Engineering Contradiction:

Improveretrieval delayVSAvoidstorage structure complexity

Core Design Contradiction:

Loss of timeVSDevice complexity

Solution Approach 1:

The patent segments the storage system into two levels: first-level storage for hot data (compression vectors and graph index) and second-level storage for cold data (original vectors). This segmentation allows frequent access operations to occur at the faster first-level storage while less frequently accessed original vectors reside in the shared storage, thereby reducing retrieval delay while maintaining adaptability to shared storage environments.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentEP4685659A1Vector retrieval methods and apparatuses, devices, and storage media

Publication Date: 2026.01.28 ALIPAY (HANGZHOU) INFORMATION TECH CO LTD

Figure 1~2
Figure 3~5
Figure 6~7

AI summary

One or more embodiments of this application provide a vector retrieval method and apparatus, a device, and a storage medium. A graph index constructed based on an original vector and a compression vector corresponding to the original vector are stored in a first-level storage resource; a node in the graph index is associated with a compression vector, and an edge represents that a near neighbor relationship exists between nodes connected to the edge; the original vector is stored in a second-level storage resource; data access performance of the first-level storage resource is better than that of the second-level storage resource; and the method includes: in response to a query vector, obtaining, from the first-level storage resource, at least one compression vector having a highest similarity with the query vector based on the graph index, and after obtaining is completed, obtaining at least one original vector corresponding to the at least one compression vector from the second-level storage resource; and separately calculating a similarity between the query vector and the at least one original vector, sorting the at least one original vector based on the calculated similarity, and determining, based on the sorting result, a vector retrieval result corresponding to the query vector.