TencentDB for PostgreSQL pgvector HNSW Performance Test Report

Download

聚焦模式

字号

最后更新时间： 2026-08-27 16:39:32

本文档由 AI 翻译

Testing Environment
Configuration Item
Description
Database
RDS PostgreSQL 17
pgvector
0.8.0
Instance Specifications
16 vCPU / 64 GB
Client
4 vCPU / 8 GB (Same VPC)
Test Tool
ann-benchmarks
Stress Test Mode
Single-threaded
Recall Metric
Recall@100
Test data set
Dataset
Level
Vector Count
Distance Type
Typical Scenario
dbpedia-openai-1M
1536
1,000,000
Cosine
OpenAI / Large Model Embedding
NYTimes-256
256
290,000
Cosine
Text similarity search
Testing Method
Preparing the Environment
1. Install ann-benchmarks.
git clone https://github.com/erikbern/ann-benchmarks.git
cd ann-benchmarks
pip install -r requirements.txt
2. Configure the database connection.
Edit ann_benchmarks/algorithms/pgvector/module.py and add the connection information:
import os
os.environ['ANN_BENCHMARKS_PG_USER'] = '<username>'
os.environ['ANN_BENCHMARKS_PG_PASSWORD'] = '<password>'
os.environ['ANN_BENCHMARKS_PG_DBNAME'] = '<database name>'
os.environ['ANN_BENCHMARKS_PG_HOST'] = '<instance private network address>'
os.environ['ANN_BENCHMARKS_PG_PORT'] = '5432'
os.environ['ANN_BENCHMARKS_PG_START_SERVICE'] = 'false'
3. Configure test parameters
Edit ann_benchmarks/algorithms/pgvector/config.yml:
float:
  any:
  - base_args: ['@metric']
    constructor: PGVector
    disabled: false
    docker_tag: ann-benchmarks-pgvector
    module: ann_benchmarks.algorithms.pgvector
    name: pgvector
    run_groups:
      M-16(200):
        arg_groups: [{M: 16, efConstruction: 200}]
        args: {}
        query_args: [[10, 20, 40, 80, 120, 200, 400, 800]]
      M-24(200):
        arg_groups: [{M: 24, efConstruction: 200}]
        args: {}
        query_args: [[10, 20, 40, 80, 120, 200, 400, 800]]
      M-48(256):
        arg_groups: [{M: 48, efConstruction: 256}]
        args: {}
        query_args: [[10, 20, 40, 80, 120, 200, 400, 800]]
4. Build the test Docker image.
python install.py --algorithm pgvector
Executing a Test
# Testing with the dbpedia-openai-1M dataset
python run.py --dataset dbpedia-openai-1000k-angular -k 100 --algorithm pgvector --runs 1
﻿
# Testing with the NYTimes-256 dataset
python run.py --dataset nytimes-256-angular -k 100 --algorithm pgvector --runs 1
Result obtaining
# Generate visualization charts
python plot.py --dataset dbpedia-openai-1000k-angular --recompute
﻿
# Export detailed CSV data
python data_export.py --out results.csv
Performance Test Results
dbpedia-openai-1M (1536 Dimensions / 1 Million Vectors)
Scenario
Index Configurations
ef_search
Recall@100
QPS
P99(ms)
High recall
m=24, ef_construction=200
120
99.2%
225
7.2
Ultra-high recall
m=48, ef_construction=256
120
99.8%
146
12.3
NYTimes-256 (256 Dimensions / 290,000 Vectors)
Scenario
Index Configurations
ef_search
Recall@100
QPS
P99(ms)
High recall
m=24, ef_construction=200
400
89.0%
101
12.7
Ultra-high recall
m=24, ef_construction=200
800
93.9%
52
23.6

帮助和支持

本页内容是否解决了您的问题？

您也可以联系销售或提交工单以寻求帮助。

填写满意度调查问卷，共创更好文档体验。

文档反馈

Configuration Item	Description
Database	RDS PostgreSQL 17
pgvector	0.8.0
Instance Specifications	16 vCPU / 64 GB
Client	4 vCPU / 8 GB (Same VPC)
Test Tool	ann-benchmarks
Stress Test Mode	Single-threaded
Recall Metric	Recall@100

Dataset	Level	Vector Count	Distance Type	Typical Scenario
dbpedia-openai-1M	1536	1,000,000	Cosine	OpenAI / Large Model Embedding
NYTimes-256	256	290,000	Cosine	Text similarity search

Scenario	Index Configurations	ef_search	Recall@100	QPS	P99(ms)
High recall	m=24, ef_construction=200	120	99.2%	225	7.2
Ultra-high recall	m=48, ef_construction=256	120	99.8%	146	12.3

tencent cloud

TencentDB for PostgreSQL

TencentDB for PostgreSQL pgvector HNSW Performance Test Report

Testing Environment

Test data set

Testing Method

Preparing the Environment

Executing a Test

Result obtaining

Performance Test Results

dbpedia-openai-1M (1536 Dimensions / 1 Million Vectors)

NYTimes-256 (256 Dimensions / 290,000 Vectors)

帮助和支持