LSG

Introduction

Local scaling graph (LSG) is a local scaling algorithm that improves the query efficiency and recall rate of HNSW indexes. This section mainly introduces the usage steps of the LSG feature of the DataVec vector engine in the openGauss database, to guide users through the operations smoothly.

**Note**

The LSG feature supports ARM/x86 architecture environments.
The LSG feature currently supports only HNSW indexes.
The LSG feature requires a certain amount of data to be inserted before building the index; otherwise, an error is reported.
The LSG feature cannot be used together with quantization methods such as PQ and RabitQ; otherwise, an error is reported.
Compatible with A/B/C/PG databases.
Supports parallel building.

Installation Preparation

Environment Requirements

The LSG feature supports ARM and x86 architecture environments.

Enabling the LSG Feature

Set the index parameter enable_lsg = on to enable the LSG feature.

Disabling the LSG Feature

Set the index parameter enable_lsg = off to disable the LSG feature.

Using LSG

HNSW-LSG

openGauss=# CREATE INDEX [INDEX_NAME] 
ON [TABLE_NAME] 
USING hnsw (COLUMN_NAME [TYPE]_[DISTANCE_FUN]_ops) 
with (m=<M>, ef_construction=<EF_CONSTRUCTION>, enable_lsg = on, lsg_degree=<REFINE_VALUE>, lsg_alpha=<REFINE_VALUE>);

INDEX_NAME - Index name
TABLE_NAME - Table name
COLUMN_NAME - Vector data column name

HNSW-LSG Index Operator

The HNSW index operator [TYPE]_[DISTANCE_FUN]_ops format:

TYPE - Vector type
- vector

The HNSW-LSG index supports the following vector data dimensions:

Name	Dimension limit
vector	2,000

DISTANCE_FUN - Distance function
- l2
- ip
- cosine

Index Operator

Index Operator	Description
vector_l2_ops	vector type - L2 distance
vector_ip_ops	vector type - inner product
vector_cosine_ops	vector type - cosine distance

Index Options

m - Maximum number of connections per layer, 2~100 (default 16)
ef_construction - Dynamic candidate set size for graph construction, 4~1000, must be greater than or equal to 2*m (default 64)
enable_lsg - Enable LSG graph construction for the HNSW index (default off)
lsg_degree - Defines the number of nearest neighbor nodes selected when calculating the isolation degree of HNSW index nodes. Type: integer, range [32, 128], default 96.
lsg_alpha - Defines the smoothness of LSG scaling for the HNSW index. Type: float, range [0, 3.0], default 2.0.
Example: Create an HNSW-LSG index using L2 distance, where the vectors in the items table are of the 2000-dimensional vector data type.
```
openGauss=# CREATE INDEX ON items USING hnsw (embedding vector_l2_ops) WITH (enable_lsg=on, lsg_degree=96, lsg_alpha=2.0);
```

Configuration Recommendations:

Other parameters are configured the same as those for the HNSW index in Vector Index.

LSG ​

Introduction ​

Installation Preparation ​

Environment Requirements ​

Enabling the LSG Feature ​

Disabling the LSG Feature ​

Using LSG ​

HNSW-LSG ​

HNSW-LSG Index Operator ​

Index Operator ​

Index Options ​

LSG