This is an automated email from the ASF dual-hosted git repository.
chengchengjin pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/incubator-gluten.git
The following commit(s) were added to refs/heads/main by this push:
new ca2ab6ad7d [DOC] Add doc about experimental feature using off-heap to
store broadcast build relation (#8882)
ca2ab6ad7d is described below
commit ca2ab6ad7d9c461b7ca1eb7b032f460ce4d567ca
Author: Terry Wang <[email protected]>
AuthorDate: Tue Mar 4 17:11:35 2025 +0800
[DOC] Add doc about experimental feature using off-heap to store broadcast
build relation (#8882)
---
docs/Configuration.md | 1 +
docs/get-started/Velox.md | 24 ++++++++++++++++++++++++
2 files changed, 25 insertions(+)
diff --git a/docs/Configuration.md b/docs/Configuration.md
index 86ee06bfb8..6c452ef86b 100644
--- a/docs/Configuration.md
+++ b/docs/Configuration.md
@@ -98,6 +98,7 @@ The following configurations are related to Velox settings.
| spark.gluten.velox.fs.s3a.connect.timeout |
Timeout for AWS s3 connection.
| 1s
|
| spark.gluten.sql.columnar.backend.velox.orc.scan.enabled |
Enable velox orc scan. If disabled, vanilla spark orc scan will be used.
| true
|
| spark.gluten.sql.complexType.scan.fallback.enabled | Force
fallback for complex type scan, including struct, map, array.
| true
|
+| spark.gluten.velox.offHeapBroadcastBuildRelation.enabled |
Experimental: If enabled, broadcast build relation will use offheap memory.
Otherwise, broadcast build relation will use onheap memory, default value is
false | |
Additionally, you can control the configurations of gluten at thread level by
local property.
diff --git a/docs/get-started/Velox.md b/docs/get-started/Velox.md
index d7e93e3f92..00cb431e17 100644
--- a/docs/get-started/Velox.md
+++ b/docs/get-started/Velox.md
@@ -545,6 +545,30 @@ I20231121 10:19:42.348845 90094332
WholeStageResultIterator.cc:220] Native Plan
queuedWallNanos sum: 2.00us, count: 1, min: 2.00us, max:
2.00us
```
+
+## Broadcast Build Relations to Off-Heap(Experimental)
+
+The experimental feature **Off-Heap Broadcast Build Relations** aims to
mitigate out-of-memory (OOM) issues caused by heap memory consumption during
broadcast operations. Detailed design
+can be found
[here](https://docs.google.com/document/d/1eZNWPUEdiz2JPJfhyVn9hrk6SqJFRNzOMZm6u5Yredk/edit?tab=t.0)
+
+### Purpose & how it works
+- **Avoid OOM**: Prevent OOM errors when broadcasting large datasets.
+- **Reduce Heap Memory Usage**: Store broadcast build relations in Spark
off-heap memory instead of on-heap memory
+
+### Configuration
+
+### Enable Off-Heap Broadcast
+To enable this feature, you can set the following Spark configuration:
+
+| Property | Default |
Description |
+|-------------------------------------------------------------|---------|-------------------------------------------------------------------|
+| `spark.gluten.velox.offHeapBroadcastBuildRelation.enabled` | `false` |
Enable/disable off-heap storage for broadcast build relations. |
+
+This feature has been tested through a series of tests, and we are collecting
more feedback from users. If you have memory problem on broadcast build
relations, please try this feature and give more feedbacks.
+
+**Note**: This feature will become the default behavior once stabilized. Stay
tuned for updates!
+
+
# Accelerators
Please refer [HBM](VeloxHBM.md) [QAT](VeloxQAT.md) [IAA](VeloxIAA.md) for
details
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
