This is an automated email from the ASF dual-hosted git repository.
yiguolei pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/doris-website.git
The following commit(s) were added to refs/heads/master by this push:
new 3d87086f8d0 Ann index performance on large scale (#3453)
3d87086f8d0 is described below
commit 3d87086f8d0c169554192314cae010eed4a1c444
Author: zhiqiang <[email protected]>
AuthorDate: Thu Mar 12 16:55:45 2026 +0800
Ann index performance on large scale (#3453)
## Versions
- [x] dev
- [x] 4.x
- [ ] 3.x
- [ ] 2.1
## Languages
- [ ] Chinese
- [ ] English
## Docs Checklist
- [ ] Checked by AI
- [ ] Test Cases Built
---
docs/ai/vector-search/performance-large-scale.md | 171 +++++++++++++++++++++
docs/ai/vector-search/performance.md | 2 +
.../ai/vector-search/performance-large-scale.md | 171 +++++++++++++++++++++
.../current/ai/vector-search/performance.md | 3 +
.../ai/vector-search/performance-large-scale.md | 171 +++++++++++++++++++++
.../version-4.x/ai/vector-search/performance.md | 3 +
.../vector-search/Performance-3BE-Import.jpg | Bin 0 -> 827176 bytes
.../Performance1536D5M-CPU-Import.png | Bin 0 -> 192519 bytes
static/images/vector-search/Performance3BE.png | Bin 0 -> 359113 bytes
.../vector-search/Performance768D-CPU-Import.png | Bin 0 -> 182689 bytes
static/images/vector-search/Performance768D10M.png | Bin 0 -> 178835 bytes
.../ai/vector-search/performance-large-scale.md | 171 +++++++++++++++++++++
.../version-4.x/ai/vector-search/performance.md | 2 +
13 files changed, 694 insertions(+)
diff --git a/docs/ai/vector-search/performance-large-scale.md
b/docs/ai/vector-search/performance-large-scale.md
new file mode 100644
index 00000000000..d9119c9a32a
--- /dev/null
+++ b/docs/ai/vector-search/performance-large-scale.md
@@ -0,0 +1,171 @@
+---
+{
+ "title": "Large-scale Performance Benchmark",
+ "language": "en",
+ "description": "Large-scale import and query benchmark results in
single-node and distributed environments for Doris ANN Index."
+}
+---
+
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+This page summarizes large-scale benchmark results in both single-node and
distributed deployments. The purpose of these tests is to show query behavior
at different data scales, and to illustrate how Doris extends vector query
capacity from single-node workloads to larger distributed deployments.
+
+## Test Matrix
+
+- Single node: FE/BE separated, BE on one 16C64GB machine.
+- Distributed: 3 BE nodes, each 16C64GB.
+- Workloads:
+ - Performance768D10M
+ - Performance1536D5M
+ - Performance768D100M
+
+## Single-node Benchmark (16C64GB)
+
+The single-node results provide a baseline for ANN query performance on
medium-to-large datasets.
+
+### Import Performance
+
+| Item | Performance768D10M | Performance1536D5M |
+|------|---------------------|--------------------|
+| Dimension | 768 | 1536 |
+| metric_type | inner_product | inner_product |
+| Rows | 10M | 5M |
+| Batch config | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch
500000` | `NUM_PER_BATCH=250000`<br/>`--stream-load-rows-per-batch 250000` |
+| Import time | 76m41s | 41m |
+| `show data all` | 56.498 GB (25.354 GB + 31.145 GB) | 55.223 GB (25.346 GB +
29.878 GB) |
+
+CPU utilization during import for Performance768D10M is shown below. The chart
indicates that CPU usage remains relatively stable throughout ingestion.
+
+<img src="/images/vector-search/Performance768D-CPU-Import.png"
alt="Performance768D10M import CPU" width="900" height="435" />
+
+For Performance1536D5M, the dataset is smaller and the batch size is also
smaller, so CPU utilization fluctuates more frequently during ingestion.
+
+<img src="/images/vector-search/Performance1536D5M-CPU-Import.png"
alt="Performance1536D5M import CPU" width="900" height="432" />
+
+### Query Performance
+
+For the two single-node workloads, Doris reaches hundreds of QPS while
maintaining high recall and low latency.
+
+#### Summary
+
+| Dataset | BestQPS | Recall@100 |
+|---------|---------|------------|
+| Performance768D10M | 481.9356 | 0.9207 |
+| Performance1536D5M | 414.7342 | 0.9677 |
+
+#### Performance768D10M (`inner_product`, 10M rows)
+
+| Concurrency | QPS | P95 Latency | P99 Latency | Avg Latency |
+|-------------|-----|-------------|-------------|-------------|
+| 10 | 116.2000 | 0.0932 | 0.0933 | 0.0861 |
+| 40 | 455.9485 | 0.1102 | 0.1225 | 0.0877 |
+| 80 | 481.9356 | 0.2331 | 0.2674 | 0.1658 |
+
+#### Performance1536D5M (`inner_product`, 5M rows)
+
+| Concurrency | QPS | P95 Latency | P99 Latency | Avg Latency |
+|-------------|-----|-------------|-------------|-------------|
+| 10 | 144.3221 | 0.0764 | 0.0800 | 0.0693 |
+| 40 | 401.9732 | 0.1271 | 0.1404 | 0.0994 |
+| 80 | 414.7342 | 0.2772 | 0.3222 | 0.1925 |
+
+In the single-node query test, the cold-query phase needs to load the full
index into memory, so CPU utilization is relatively low while the system waits
for IO. During the warm-query phase, CPU utilization increases significantly
and approaches 100%.
+
+<img src="/images/vector-search/Performance768D10M.png"
alt="Performance768D10M query CPU" width="900" height="430" />
+
+## Distributed Benchmark (3 x 16C64GB)
+
+The distributed test focuses on a larger dataset that exceeds the practical
memory envelope of a single 16C64GB node.
+
+For 3BE testing, `Performance768D100M` was selected. Since single-node memory
is limited to 64GB, vector quantization is enabled to reduce memory usage. This
test is intended to show how Doris sustains vector query capability at 100M
scale through multi-BE deployment, rather than to provide a direct one-to-one
comparison with the smaller single-node cases.
+
+### Import Performance
+
+| Item | Value |
+|------|-------|
+| Dataset | Performance768D100M |
+| Rows | 100M |
+| Dimension | 768 |
+| Batch config | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch
500000` |
+| Index properties | `"dim"="768", "index_type"="hnsw",
"metric_type"="l2_distance", "pq_m"="384", "pq_nbits"="8", "quantizer"="pq"` |
+| Build index time | 4h5min |
+| `show data all` | 198.809 GB (137.259 GB + 61.550 GB) |
+
+Post-build distribution:
+
+- 3 buckets
+- 34 rowsets per bucket, each rowset about 1.99 GB
+- 6 segments per rowset
+
+### Query Performance
+
+#### Summary
+
+| Metric | Value |
+|--------|-------|
+| BestQPS | 77.6247 |
+| Recall@100 | 0.9294 |
+
+#### Detailed results (`l2_distance`, 100M rows)
+
+| Concurrency | QPS | P95 Latency | P99 Latency | Avg Latency |
+|-------------|-----|-------------|-------------|-------------|
+| 10 | 46.5836 | 0.2628 | 0.2791 | 0.2145 |
+| 20 | 75.3579 | 0.3251 | 0.3541 | 0.2651 |
+| 30 | 77.6247 | 0.5222 | 0.5766 | 0.3860 |
+| 40 | 76.6313 | 0.7089 | 0.7854 | 0.5212 |
+
+During index build, CPU utilization stays around 50%, indicating that the
build process does not saturate CPU resources for an extended period.
+
+<img src="/images/vector-search/Performance-3BE-Import.jpg"
alt="Performance768D100M import CPU" width="900" height="444" />
+
+The chart below shows CPU utilization during the query phase. CPU usage stays
at a relatively high level across the nodes, indicating that the distributed
query workload makes good use of available compute resources.
+
+<img src="/images/vector-search/Performance3BE.png" alt="Performance768D100M
query CPU" width="900" height="323" />
+
+## Summary
+
+- On tens of millions of vectors, Doris provides strong ANN query performance
on a single node, with hundreds of QPS and high recall.
+- On a 100M-vector dataset, Doris continues to provide online vector query
capability through multi-BE deployment.
+- Because the test groups use different dataset sizes, distance metrics, and
index settings, the results should be read as scale benchmarks rather than
direct one-to-one performance comparisons.
+
+## Notes
+
+- Metric types differ between the two test groups (`inner_product` vs
`l2_distance`), so absolute values should not be compared directly.
+- The single-node `Performance768D10M` result at concurrency = 10 has been
adjusted to exclude cold-query impact.
+
+## Reproduction
+
+Single-node:
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D10M
--stream-load-rows-per-batch 500000
+
+export NUM_PER_BATCH=250000
+vectordbbench doris ... --case-type Performance1536D5M
--stream-load-rows-per-batch 250000
+```
+
+Distributed 3BE:
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D100M
--stream-load-rows-per-batch 500000
+```
diff --git a/docs/ai/vector-search/performance.md
b/docs/ai/vector-search/performance.md
index fa0ccf4aafd..a0b669c5658 100644
--- a/docs/ai/vector-search/performance.md
+++ b/docs/ai/vector-search/performance.md
@@ -27,6 +27,8 @@ under the License.
This document introduces the query and ingestion performance of Doris ANN
Index. All benchmarks are conducted using
[VectorDBBench](https://github.com/zilliztech/VectorDBBench).
+For larger-scale results (10M/100M-level datasets) in single-node and
distributed setups, see [Large-scale Performance
Benchmark](./performance-large-scale.md).
+
## Test Environment
All machines used in the tests have 16 CPU cores and 64 GB memory, with CPUs
of model Intel(R) Xeon(R) Platinum 8369B CPU @ 2.70GHz. FE and BE are
co-located on a single 16C64GB machine (note: this is not a recommended
production deployment; in production, FE and BE should be deployed separately).
The tested version is Apache Doris 4.0.2. The dataset used is VectorDBBench
Performance768D1M, which contains 1 million vectors of 768 dimensions.
diff --git
a/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance-large-scale.md
b/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance-large-scale.md
new file mode 100644
index 00000000000..1e741df75c0
--- /dev/null
+++
b/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance-large-scale.md
@@ -0,0 +1,171 @@
+---
+{
+ "title": "大规模性能实测",
+ "language": "zh-CN",
+ "description": "总结 Doris ANN Index 在单机与分布式环境下的大规模导入与查询性能测试结果。"
+}
+---
+
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+本文总结了在单机和分布式两种部署下的大规模测试结果。测试的目的，是展示 Doris
在不同数据规模下的查询表现，以及在数据规模持续增长时，如何将向量查询能力从单机扩展到分布式部署。
+
+## 测试矩阵
+
+- 单机：FE / BE 分离部署，BE 使用 1 台 16C64GB 机器。
+- 分布式：3 台 BE，每台 16C64GB。
+- 测试数据集：
+ - Performance768D10M
+ - Performance1536D5M
+ - Performance768D100M
+
+## 单机实测（16C64GB）
+
+单机结果给出了中大规模数据集上的 ANN 查询性能基线。
+
+### 导入性能
+
+| 项目 | Performance768D10M | Performance1536D5M |
+|------|---------------------|--------------------|
+| 向量维度 | 768 | 1536 |
+| metric_type | inner_product | inner_product |
+| 数据量 | 10M 行 | 5M 行 |
+| 导入 batch 参数 | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch
500000` | `NUM_PER_BATCH=250000`<br/>`--stream-load-rows-per-batch 250000` |
+| 导入耗时 | 76m41s | 41m |
+| `show data all` | 56.498 GB (25.354 GB + 31.145 GB) | 55.223 GB (25.346 GB +
29.878 GB) |
+
+Performance768D10M 导入过程中的 CPU 监控如下。可以看到，导入期间 CPU 使用率整体较为平稳。
+
+<img src="/images/vector-search/Performance768D-CPU-Import.png"
alt="Performance768D10M import CPU" width="900" height="435" />
+
+Performance1536D5M 的数据量较小，导入时的 batch size 也更小，因此导入阶段的 CPU 使用率波动更为频繁。
+
+<img src="/images/vector-search/Performance1536D5M-CPU-Import.png"
alt="Performance1536D5M import CPU" width="900" height="432" />
+
+### 查询性能
+
+从两个单机 workload 的结果可以看到，Doris 在保持较高召回率的同时，能够达到数百 QPS，并维持较低查询延迟。
+
+#### 汇总
+
+| 数据集 | BestQPS | Recall@100 |
+|--------|---------|------------|
+| Performance768D10M | 481.9356 | 0.9207 |
+| Performance1536D5M | 414.7342 | 0.9677 |
+
+#### Performance768D10M（`inner_product`, 10M 行）
+
+| 并发数 | QPS | P95 延迟 | P99 延迟 | 平均延迟 |
+|--------|-----|----------|----------|----------|
+| 10 | 116.2000 | 0.0932 | 0.0933 | 0.0861 |
+| 40 | 455.9485 | 0.1102 | 0.1225 | 0.0877 |
+| 80 | 481.9356 | 0.2331 | 0.2674 | 0.1658 |
+
+#### Performance1536D5M（`inner_product`, 5M 行）
+
+| 并发数 | QPS | P95 延迟 | P99 延迟 | 平均延迟 |
+|--------|-----|----------|----------|----------|
+| 10 | 144.3221 | 0.0764 | 0.0800 | 0.0693 |
+| 40 | 401.9732 | 0.1271 | 0.1404 | 0.0994 |
+| 80 | 414.7342 | 0.2772 | 0.3222 | 0.1925 |
+
+在单机查询场景下，冷查询阶段需要将索引加载到内存，因此 CPU 利用率相对较低，系统主要在等待 IO 完成；进入热查询阶段后，CPU 利用率明显提升并接近
100%。
+
+<img src="/images/vector-search/Performance768D10M.png"
alt="Performance768D10M query CPU" width="900" height="430" />
+
+## 分布式实测（3 × 16C64GB）
+
+分布式测试聚焦于更大规模的数据集，该数据规模已经超出单台 16C64GB 机器较为合适的内存承载范围。
+
+3BE 场景使用 `Performance768D100M` 数据集。由于单机内存上限为 64GB，采用向量量化压缩以降低内存开销。该测试的重点，是展示
Doris 在 100M 规模下如何通过多 BE 部署继续提供向量查询能力，而不是与单机小规模场景做一一对应的绝对数值比较。
+
+### 导入性能
+
+| 项目 | 数值 |
+|------|------|
+| 数据集 | Performance768D100M |
+| 数据量 | 100M 行 |
+| 维度 | 768 |
+| batch 参数 | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch 500000` |
+| 索引参数 | `"dim"="768", "index_type"="hnsw", "metric_type"="l2_distance",
"pq_m"="384", "pq_nbits"="8", "quantizer"="pq"` |
+| build index 用时 | 4h5min |
+| `show data all` | 198.809 GB (137.259 GB + 61.550 GB) |
+
+build index 后的数据分布：
+
+- 3 个 bucket
+- 每个 bucket 34 个 rowset，每个 rowset 约 1.99 GB
+- 每个 rowset 6 个 segment
+
+build index 期间 CPU 使用率整体稳定在约 50%，说明构建过程未长时间打满 CPU，仍保留了一定资源余量。
+
+<img src="/images/vector-search/Performance-3BE-Import.jpg"
alt="Performance768D100M import CPU" width="900" height="444" />
+
+### 查询性能
+
+#### 汇总
+
+| 指标 | 数值 |
+|------|------|
+| BestQPS | 77.6247 |
+| Recall@100 | 0.9294 |
+
+#### 明细（`l2_distance`, 100M 行）
+
+| 并发数 | QPS | P95 延迟 | P99 延迟 | 平均延迟 |
+|--------|-----|----------|----------|----------|
+| 10 | 46.5836 | 0.2628 | 0.2791 | 0.2145 |
+| 20 | 75.3579 | 0.3251 | 0.3541 | 0.2651 |
+| 30 | 77.6247 | 0.5222 | 0.5766 | 0.3860 |
+| 40 | 76.6313 | 0.7089 | 0.7854 | 0.5212 |
+
+下图为查询阶段的 CPU 监控，可以看到各节点 CPU 使用率保持在较高水平，说明查询负载较充分地利用了分布式计算资源。
+
+<img src="/images/vector-search/Performance3BE.png" alt="Performance768D100M
query CPU" width="900" height="323" />
+
+## 总结
+
+- 在千万级向量数据规模下，Doris 在单机场景能够提供较强的 ANN 查询性能，达到数百 QPS，并保持较高召回率。
+- 在 100M 向量数据集上，Doris 可通过多 BE 部署继续提供在线向量查询能力。
+- 由于不同测试组采用了不同的数据规模、距离度量和索引参数，这些结果更适合用于观察规模扩展表现，而不适合做一一对应的绝对数值对比。
+
+## 说明
+
+- 两组测试的距离度量不同（`inner_product` 与 `l2_distance`），不建议直接横向对比绝对数值。
+- 单机 `Performance768D10M` 在并发 10 下的结果已剔除冷查询影响后进行修正。
+
+## 复现方式
+
+单机：
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D10M
--stream-load-rows-per-batch 500000
+
+export NUM_PER_BATCH=250000
+vectordbbench doris ... --case-type Performance1536D5M
--stream-load-rows-per-batch 250000
+```
+
+分布式 3BE：
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D100M
--stream-load-rows-per-batch 500000
+```
diff --git
a/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance.md
b/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance.md
index f9acb20ac69..c7d591a6e06 100644
---
a/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance.md
+++
b/i18n/zh-CN/docusaurus-plugin-content-docs/current/ai/vector-search/performance.md
@@ -26,6 +26,9 @@ under the License.
-->
这里介绍 Doris Ann Index 的查询性能与导入性能。所有的性能测试都是通过
[VectorDBBench](https://github.com/zilliztech/VectorDBBench) 完成的。
+
+如需查看单机与分布式下更大规模（10M/100M 级）数据集的测试结果，请参考[大规模性能实测](./performance-large-scale.md)。
+
## 测试环境
测试涉及到的机器规格都是 16C64GB的机器，CPU型号为 Intel(R) Xeon(R) Platinum 8369B CPU @
2.70GHz。FE BE 混合部署在一台 16C64GB 的机器上（注意这个不是推荐的生产环境部署方式，生产环境上建议 FE 与 BE
分离部署），测试的版本是 Apache Doris 4.0.2。采用的测试数据集是 VectorDBBench
Performance768D1M，该数据集向量的维度为 768 维，一共一百万行。
diff --git
a/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance-large-scale.md
b/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance-large-scale.md
new file mode 100644
index 00000000000..d6847a544d8
--- /dev/null
+++
b/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance-large-scale.md
@@ -0,0 +1,171 @@
+---
+{
+ "title": "大规模性能实测",
+ "language": "zh-CN",
+ "description": "总结 Doris ANN Index 在单机与分布式环境下的大规模导入与查询性能测试结果。"
+}
+---
+
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+本文总结了在单机和分布式两种部署下的大规模测试结果。测试的目的，是展示 Doris
在不同数据规模下的查询表现，以及在数据规模持续增长时，如何将向量查询能力从单机扩展到分布式部署。
+
+## 测试矩阵
+
+- 单机：FE / BE 分离部署，BE 使用 1 台 16C64GB 机器。
+- 分布式：3 台 BE，每台 16C64GB。
+- 测试数据集：
+ - Performance768D10M
+ - Performance1536D5M
+ - Performance768D100M
+
+## 单机实测（16C64GB）
+
+单机结果给出了中大规模数据集上的 ANN 查询性能基线。
+
+### 导入性能
+
+| 项目 | Performance768D10M | Performance1536D5M |
+|------|---------------------|--------------------|
+| 向量维度 | 768 | 1536 |
+| metric_type | inner_product | inner_product |
+| 数据量 | 10M 行 | 5M 行 |
+| 导入 batch 参数 | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch
500000` | `NUM_PER_BATCH=250000`<br/>`--stream-load-rows-per-batch 250000` |
+| 导入耗时 | 76m41s | 41m |
+| `show data all` | 56.498 GB (25.354 GB + 31.145 GB) | 55.223 GB (25.346 GB +
29.878 GB) |
+
+Performance768D10M 导入过程中的 CPU 监控如下。可以看到，导入期间 CPU 使用率整体较为平稳。
+
+<img src="/images/vector-search/Performance768D-CPU-Import.png"
alt="Performance768D10M import CPU" width="900" height="435" />
+
+Performance1536D5M 的数据量较小，导入时的 batch size 也更小，因此导入阶段的 CPU 使用率波动更为频繁。
+
+<img src="/images/vector-search/Performance1536D5M-CPU-Import.png"
alt="Performance1536D5M import CPU" width="900" height="432" />
+
+### 查询性能
+
+从两个单机 workload 的结果可以看到，Doris 在保持较高召回率的同时，能够达到数百 QPS，并维持较低查询延迟。
+
+#### 汇总
+
+| 数据集 | BestQPS | Recall@100 |
+|--------|---------|------------|
+| Performance768D10M | 481.9356 | 0.9207 |
+| Performance1536D5M | 414.7342 | 0.9677 |
+
+#### Performance768D10M（`inner_product`, 10M 行）
+
+| 并发数 | QPS | P95 延迟 | P99 延迟 | 平均延迟 |
+|--------|-----|----------|----------|----------|
+| 10 | 116.2000 | 0.0932 | 0.0933 | 0.0861 |
+| 40 | 455.9485 | 0.1102 | 0.1225 | 0.0877 |
+| 80 | 481.9356 | 0.2331 | 0.2674 | 0.1658 |
+
+#### Performance1536D5M（`inner_product`, 5M 行）
+
+| 并发数 | QPS | P95 延迟 | P99 延迟 | 平均延迟 |
+|--------|-----|----------|----------|----------|
+| 10 | 144.3221 | 0.0764 | 0.0800 | 0.0693 |
+| 40 | 401.9732 | 0.1271 | 0.1404 | 0.0994 |
+| 80 | 414.7342 | 0.2772 | 0.3222 | 0.1925 |
+
+在单机查询场景下，冷查询阶段需要将索引加载到内存，因此 CPU 利用率相对较低，系统主要在等待 IO 完成；进入热查询阶段后，CPU 利用率明显提升并接近
100%。
+
+<img src="/images/vector-search/Performance768D10M.png"
alt="Performance768D10M query CPU" width="900" height="430" />
+
+## 分布式实测（3 × 16C64GB）
+
+分布式测试聚焦于更大规模的数据集，该数据规模已经超出单台 16C64GB 机器较为合适的内存承载范围。
+
+3BE 场景使用 `Performance768D100M` 数据集。由于单机内存上限为 64GB，采用向量量化压缩以降低内存开销。该测试的重点，是展示
Doris 在 100M 规模下如何通过多 BE 部署继续提供向量查询能力，而不是与单机小规模场景做一一对应的绝对数值比较。
+
+### 导入性能
+
+| 项目 | 数值 |
+|------|------|
+| 数据集 | Performance768D100M |
+| 数据量 | 100M 行 |
+| 维度 | 768 |
+| batch 参数 | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch 500000` |
+| 索引参数 | `"dim"="768", "index_type"="hnsw", "metric_type"="l2_distance",
"pq_m"="384", "pq_nbits"="8", "quantizer"="pq"` |
+| build index 用时 | 4h5min |
+| `show data all` | 198.809 GB (137.259 GB + 61.550 GB) |
+
+build index 后的数据分布：
+
+- 3 个 bucket
+- 每个 bucket 34 个 rowset，每个 rowset 约 1.99 GB
+- 每个 rowset 6 个 segment
+
+### 查询性能
+
+#### 汇总
+
+| 指标 | 数值 |
+|------|------|
+| BestQPS | 77.6247 |
+| Recall@100 | 0.9294 |
+
+#### 明细（`l2_distance`, 100M 行）
+
+| 并发数 | QPS | P95 延迟 | P99 延迟 | 平均延迟 |
+|--------|-----|----------|----------|----------|
+| 10 | 46.5836 | 0.2628 | 0.2791 | 0.2145 |
+| 20 | 75.3579 | 0.3251 | 0.3541 | 0.2651 |
+| 30 | 77.6247 | 0.5222 | 0.5766 | 0.3860 |
+| 40 | 76.6313 | 0.7089 | 0.7854 | 0.5212 |
+
+build index 期间 CPU 使用率整体稳定在约 50%，说明构建过程未长时间打满 CPU，仍保留了一定资源余量。
+
+<img src="/images/vector-search/Performance-3BE-Import.jpg"
alt="Performance768D100M import CPU" width="900" height="444" />
+
+下图为查询阶段的 CPU 监控，可以看到各节点 CPU 使用率保持在较高水平，说明查询负载较充分地利用了分布式计算资源。
+
+<img src="/images/vector-search/Performance3BE.png" alt="Performance768D100M
query CPU" width="900" height="323" />
+
+## 总结
+
+- 在千万级向量数据规模下，Doris 在单机场景能够提供较强的 ANN 查询性能，达到数百 QPS，并保持较高召回率。
+- 在 100M 向量数据集上，Doris 可通过多 BE 部署继续提供在线向量查询能力。
+- 由于不同测试组采用了不同的数据规模、距离度量和索引参数，这些结果更适合用于观察规模扩展表现，而不适合做一一对应的绝对数值对比。
+
+## 说明
+
+- 两组测试的距离度量不同（`inner_product` 与 `l2_distance`），不建议直接横向对比绝对数值。
+- 单机 `Performance768D10M` 在并发 10 下的结果已剔除冷查询影响后进行修正。
+
+## 复现方式
+
+单机：
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D10M
--stream-load-rows-per-batch 500000
+
+export NUM_PER_BATCH=250000
+vectordbbench doris ... --case-type Performance1536D5M
--stream-load-rows-per-batch 250000
+```
+
+分布式 3BE：
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D100M
--stream-load-rows-per-batch 500000
+```
diff --git
a/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance.md
b/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance.md
index ff45714af84..de57eb37887 100644
---
a/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance.md
+++
b/i18n/zh-CN/docusaurus-plugin-content-docs/version-4.x/ai/vector-search/performance.md
@@ -26,6 +26,9 @@ under the License.
-->
这里介绍 Doris Ann Index 的查询性能与导入性能。所有的性能测试都是通过
[VectorDBBench](https://github.com/zilliztech/VectorDBBench) 完成的。
+
+如需查看单机与分布式下更大规模（10M/100M 级）数据集的测试结果，请参考[大规模性能实测](./performance-large-scale.md)。
+
## 测试环境
测试涉及到的机器规格都是 16C64GB 的机器，CPU型号为 Intel(R) Xeon(R) Platinum 8369B CPU @
2.70GHz。FE BE 混合部署在一台 16C64GB 的机器上（注意这个不是推荐的生产环境部署方式，生产环境上建议 FE 与 BE
分离部署），测试的版本是 Apache Doris 4.0.2。采用的测试数据集是 VectorDBBench
Performance768D1M，该数据集向量的维度为 768 维，一共一百万行。
diff --git a/static/images/vector-search/Performance-3BE-Import.jpg
b/static/images/vector-search/Performance-3BE-Import.jpg
new file mode 100644
index 00000000000..4ab3eb48e73
Binary files /dev/null and
b/static/images/vector-search/Performance-3BE-Import.jpg differ
diff --git a/static/images/vector-search/Performance1536D5M-CPU-Import.png
b/static/images/vector-search/Performance1536D5M-CPU-Import.png
new file mode 100644
index 00000000000..f626efdcbfe
Binary files /dev/null and
b/static/images/vector-search/Performance1536D5M-CPU-Import.png differ
diff --git a/static/images/vector-search/Performance3BE.png
b/static/images/vector-search/Performance3BE.png
new file mode 100644
index 00000000000..3c412ee1a6b
Binary files /dev/null and b/static/images/vector-search/Performance3BE.png
differ
diff --git a/static/images/vector-search/Performance768D-CPU-Import.png
b/static/images/vector-search/Performance768D-CPU-Import.png
new file mode 100644
index 00000000000..fabe13e7768
Binary files /dev/null and
b/static/images/vector-search/Performance768D-CPU-Import.png differ
diff --git a/static/images/vector-search/Performance768D10M.png
b/static/images/vector-search/Performance768D10M.png
new file mode 100644
index 00000000000..752a16af4b8
Binary files /dev/null and b/static/images/vector-search/Performance768D10M.png
differ
diff --git
a/versioned_docs/version-4.x/ai/vector-search/performance-large-scale.md
b/versioned_docs/version-4.x/ai/vector-search/performance-large-scale.md
new file mode 100644
index 00000000000..d9119c9a32a
--- /dev/null
+++ b/versioned_docs/version-4.x/ai/vector-search/performance-large-scale.md
@@ -0,0 +1,171 @@
+---
+{
+ "title": "Large-scale Performance Benchmark",
+ "language": "en",
+ "description": "Large-scale import and query benchmark results in
single-node and distributed environments for Doris ANN Index."
+}
+---
+
+<!--
+Licensed to the Apache Software Foundation (ASF) under one
+or more contributor license agreements. See the NOTICE file
+distributed with this work for additional information
+regarding copyright ownership. The ASF licenses this file
+to you under the Apache License, Version 2.0 (the
+"License"); you may not use this file except in compliance
+with the License. You may obtain a copy of the License at
+
+ http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing,
+software distributed under the License is distributed on an
+"AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+KIND, either express or implied. See the License for the
+specific language governing permissions and limitations
+under the License.
+-->
+
+This page summarizes large-scale benchmark results in both single-node and
distributed deployments. The purpose of these tests is to show query behavior
at different data scales, and to illustrate how Doris extends vector query
capacity from single-node workloads to larger distributed deployments.
+
+## Test Matrix
+
+- Single node: FE/BE separated, BE on one 16C64GB machine.
+- Distributed: 3 BE nodes, each 16C64GB.
+- Workloads:
+ - Performance768D10M
+ - Performance1536D5M
+ - Performance768D100M
+
+## Single-node Benchmark (16C64GB)
+
+The single-node results provide a baseline for ANN query performance on
medium-to-large datasets.
+
+### Import Performance
+
+| Item | Performance768D10M | Performance1536D5M |
+|------|---------------------|--------------------|
+| Dimension | 768 | 1536 |
+| metric_type | inner_product | inner_product |
+| Rows | 10M | 5M |
+| Batch config | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch
500000` | `NUM_PER_BATCH=250000`<br/>`--stream-load-rows-per-batch 250000` |
+| Import time | 76m41s | 41m |
+| `show data all` | 56.498 GB (25.354 GB + 31.145 GB) | 55.223 GB (25.346 GB +
29.878 GB) |
+
+CPU utilization during import for Performance768D10M is shown below. The chart
indicates that CPU usage remains relatively stable throughout ingestion.
+
+<img src="/images/vector-search/Performance768D-CPU-Import.png"
alt="Performance768D10M import CPU" width="900" height="435" />
+
+For Performance1536D5M, the dataset is smaller and the batch size is also
smaller, so CPU utilization fluctuates more frequently during ingestion.
+
+<img src="/images/vector-search/Performance1536D5M-CPU-Import.png"
alt="Performance1536D5M import CPU" width="900" height="432" />
+
+### Query Performance
+
+For the two single-node workloads, Doris reaches hundreds of QPS while
maintaining high recall and low latency.
+
+#### Summary
+
+| Dataset | BestQPS | Recall@100 |
+|---------|---------|------------|
+| Performance768D10M | 481.9356 | 0.9207 |
+| Performance1536D5M | 414.7342 | 0.9677 |
+
+#### Performance768D10M (`inner_product`, 10M rows)
+
+| Concurrency | QPS | P95 Latency | P99 Latency | Avg Latency |
+|-------------|-----|-------------|-------------|-------------|
+| 10 | 116.2000 | 0.0932 | 0.0933 | 0.0861 |
+| 40 | 455.9485 | 0.1102 | 0.1225 | 0.0877 |
+| 80 | 481.9356 | 0.2331 | 0.2674 | 0.1658 |
+
+#### Performance1536D5M (`inner_product`, 5M rows)
+
+| Concurrency | QPS | P95 Latency | P99 Latency | Avg Latency |
+|-------------|-----|-------------|-------------|-------------|
+| 10 | 144.3221 | 0.0764 | 0.0800 | 0.0693 |
+| 40 | 401.9732 | 0.1271 | 0.1404 | 0.0994 |
+| 80 | 414.7342 | 0.2772 | 0.3222 | 0.1925 |
+
+In the single-node query test, the cold-query phase needs to load the full
index into memory, so CPU utilization is relatively low while the system waits
for IO. During the warm-query phase, CPU utilization increases significantly
and approaches 100%.
+
+<img src="/images/vector-search/Performance768D10M.png"
alt="Performance768D10M query CPU" width="900" height="430" />
+
+## Distributed Benchmark (3 x 16C64GB)
+
+The distributed test focuses on a larger dataset that exceeds the practical
memory envelope of a single 16C64GB node.
+
+For 3BE testing, `Performance768D100M` was selected. Since single-node memory
is limited to 64GB, vector quantization is enabled to reduce memory usage. This
test is intended to show how Doris sustains vector query capability at 100M
scale through multi-BE deployment, rather than to provide a direct one-to-one
comparison with the smaller single-node cases.
+
+### Import Performance
+
+| Item | Value |
+|------|-------|
+| Dataset | Performance768D100M |
+| Rows | 100M |
+| Dimension | 768 |
+| Batch config | `NUM_PER_BATCH=500000`<br/>`--stream-load-rows-per-batch
500000` |
+| Index properties | `"dim"="768", "index_type"="hnsw",
"metric_type"="l2_distance", "pq_m"="384", "pq_nbits"="8", "quantizer"="pq"` |
+| Build index time | 4h5min |
+| `show data all` | 198.809 GB (137.259 GB + 61.550 GB) |
+
+Post-build distribution:
+
+- 3 buckets
+- 34 rowsets per bucket, each rowset about 1.99 GB
+- 6 segments per rowset
+
+### Query Performance
+
+#### Summary
+
+| Metric | Value |
+|--------|-------|
+| BestQPS | 77.6247 |
+| Recall@100 | 0.9294 |
+
+#### Detailed results (`l2_distance`, 100M rows)
+
+| Concurrency | QPS | P95 Latency | P99 Latency | Avg Latency |
+|-------------|-----|-------------|-------------|-------------|
+| 10 | 46.5836 | 0.2628 | 0.2791 | 0.2145 |
+| 20 | 75.3579 | 0.3251 | 0.3541 | 0.2651 |
+| 30 | 77.6247 | 0.5222 | 0.5766 | 0.3860 |
+| 40 | 76.6313 | 0.7089 | 0.7854 | 0.5212 |
+
+During index build, CPU utilization stays around 50%, indicating that the
build process does not saturate CPU resources for an extended period.
+
+<img src="/images/vector-search/Performance-3BE-Import.jpg"
alt="Performance768D100M import CPU" width="900" height="444" />
+
+The chart below shows CPU utilization during the query phase. CPU usage stays
at a relatively high level across the nodes, indicating that the distributed
query workload makes good use of available compute resources.
+
+<img src="/images/vector-search/Performance3BE.png" alt="Performance768D100M
query CPU" width="900" height="323" />
+
+## Summary
+
+- On tens of millions of vectors, Doris provides strong ANN query performance
on a single node, with hundreds of QPS and high recall.
+- On a 100M-vector dataset, Doris continues to provide online vector query
capability through multi-BE deployment.
+- Because the test groups use different dataset sizes, distance metrics, and
index settings, the results should be read as scale benchmarks rather than
direct one-to-one performance comparisons.
+
+## Notes
+
+- Metric types differ between the two test groups (`inner_product` vs
`l2_distance`), so absolute values should not be compared directly.
+- The single-node `Performance768D10M` result at concurrency = 10 has been
adjusted to exclude cold-query impact.
+
+## Reproduction
+
+Single-node:
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D10M
--stream-load-rows-per-batch 500000
+
+export NUM_PER_BATCH=250000
+vectordbbench doris ... --case-type Performance1536D5M
--stream-load-rows-per-batch 250000
+```
+
+Distributed 3BE:
+
+```bash
+export NUM_PER_BATCH=500000
+vectordbbench doris ... --case-type Performance768D100M
--stream-load-rows-per-batch 500000
+```
diff --git a/versioned_docs/version-4.x/ai/vector-search/performance.md
b/versioned_docs/version-4.x/ai/vector-search/performance.md
index fa0ccf4aafd..a0b669c5658 100644
--- a/versioned_docs/version-4.x/ai/vector-search/performance.md
+++ b/versioned_docs/version-4.x/ai/vector-search/performance.md
@@ -27,6 +27,8 @@ under the License.
This document introduces the query and ingestion performance of Doris ANN
Index. All benchmarks are conducted using
[VectorDBBench](https://github.com/zilliztech/VectorDBBench).
+For larger-scale results (10M/100M-level datasets) in single-node and
distributed setups, see [Large-scale Performance
Benchmark](./performance-large-scale.md).
+
## Test Environment
All machines used in the tests have 16 CPU cores and 64 GB memory, with CPUs
of model Intel(R) Xeon(R) Platinum 8369B CPU @ 2.70GHz. FE and BE are
co-located on a single 16C64GB machine (note: this is not a recommended
production deployment; in production, FE and BE should be deployed separately).
The tested version is Apache Doris 4.0.2. The dataset used is VectorDBBench
Performance768D1M, which contains 1 million vectors of 768 dimensions.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
