US 2025/0094400 A1 · DEPLOYING A VECTOR INDEX ON MULTIPLE NODES OF A CLUSTER

Abstract

Techniques for deploying a vector index on multiple nodes of a cluster are provided. In one technique, an instruction is received to create a vector index on a set of vectors that is stored in a vector database that is connected to the multiple nodes. In response, an HNSW index is created based on the set of vectors and the HNSW index is stored on each node. In response to receiving a vector query, a node processes the vector query against its copy of the HNSW index. In another technique, each node retrieves, from a vector database, a respective subset of a set of vectors and generates, based on the respective subset, a respective HNSW index. A vector query is transmitted to each node, which traverses its HNSW index to generate results of the vector query. The results from each node are combined to generate final results.

Claims20

Dependent claims

A method comprising:

receiving an instruction to create a vector index on a set of vectors that is stored in a vector database that is communicatively coupled to a plurality of nodes in a cluster;
in response to receiving the instruction:
creating an HNSW index based on the set of vectors;
causing a copy of the HNSW index to be stored in memory of each node in the plurality of nodes;
in response to receiving a first vector query, processing, by a first node in the plurality of nodes, the first vector query against a first copy of the HNSW index;
in response to receiving a second vector query, processing, by a second node in the plurality of nodes, the second vector query against a second copy of the HNSW index;
wherein the method is performed by one or more computing devices.

4 more elements

One or more non-transitory storage media storing instructions which, when executed by one or more computing devices, cause:

receiving an instruction to create a vector index on a set of vectors that is stored in a vector database that is communicatively coupled to a plurality of nodes in a cluster;
in response to receiving the instruction:
creating an HNSW index based on the set of vectors;
causing a copy of the HNSW index to be stored in memory of each node in the plurality of nodes;
in response to receiving a first vector query, processing, by a first node in the plurality of nodes, the first vector query against a first copy of the HNSW index;
in response to receiving a second vector query, processing, by a second node in the plurality of nodes, the second vector query against a second copy of the HNSW index.

3 more elements

Dependencies are inferred from the claim text.

Deploying a vector index on multiple nodes of a cluster

Life of this patent

Deploying a vector index on multiple nodes of a cluster

Life of this patent

Abstract

Claims20