[
https://issues.apache.org/jira/browse/SPARK-34336?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17277485#comment-17277485
]
Baohe Zhang commented on SPARK-34336:
-------------------------------------
Full benchmark results are added as txt attachments.
> Use GenericData as Avro serialization data model can improve Avro write/read
> performance
> ----------------------------------------------------------------------------------------
>
> Key: SPARK-34336
> URL: https://issues.apache.org/jira/browse/SPARK-34336
> Project: Spark
> Issue Type: Improvement
> Components: Input/Output, SQL
> Affects Versions: 3.1.2
> Reporter: Baohe Zhang
> Priority: Major
> Attachments: base_read.txt, base_write.txt, generic_data_read.txt,
> generic_data_write.txt, read_comparison.png, write_comparison.png
>
>
> We found that using "org.apache.avro.generic.GenericData" as Avro
> serialization data model in Avro writer can significantly improve Avro write
> performance and slightly improve Avro read performance.
> This optimization was originally put up by [~samkhan] in this PR
> https://github.com/apache/spark/pull/29354.
> We re-evaluated the change "Use GenericData instead of ReflectData when
> writing Avro data" in that PR and verified it can provide performance
> improvement in Avro write/read benchmarks.
> The base branch is today(2/2/21)'s branch-3.1.
> Besides current Avro read/write benchmarks, I also ran some extra benchmarks
> for nested structs and arrays read/write, these benchmarks were put up in
> this PR https://github.com/apache/spark/pull/29352 but haven't been merged.
> Benchmark results are added in the comment.
--
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
