[
https://issues.apache.org/jira/browse/SPARK-32184?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yuan Zhou updated SPARK-32184:
------------------------------
Summary: Performance regression on TPCH Q18 in 3.0 - ~20% slower than 2.4
(was: Performance regression on TPCH Q18)
> Performance regression on TPCH Q18 in 3.0 - ~20% slower than 2.4
> ----------------------------------------------------------------
>
> Key: SPARK-32184
> URL: https://issues.apache.org/jira/browse/SPARK-32184
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 3.0.0
> Environment: spark 2.4 and spark 3.0 are using the same configurations
> * spark.driver.memory 20g
> * spark.executor.memory 20g
> * spark.executor.cores 7
> * spark.executor.memoryOverhead 3g
> * spark.sql.shuffle.partitions 384
> Reporter: Yuan Zhou
> Priority: Major
> Attachments: 2.4.png, 3.0.png
>
>
> Hi Spark developers,
> Testing with the new Spark 3.0.0 here and found some performance regression
> on TPCH Q18. Spark 2.4 seems can "reuse" the HashAgg results in two SMJ,
> while Spark 3.0.0 needs to calculate this results twice. I tried to search
> the documentation but no turnings found on this.
> Here's the SQL diagram for 2.4. The 2nd exchange on the top right corner was
> reused. !2.4.png!
> Here's the diagram for 3.0. The 1st exchange on top-right corner was reused.
> !3.0.png!
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
