[jira] [Commented] (SPARK-16073) Performance of Parquet encodings on saving primitive arrays

["Kazuaki Ishizaki (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-16073?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15740807#comment-15740807 ] 

Kazuaki Ishizaki commented on SPARK-16073:
------------------------------------------

It is an interesting topic. In the current situation, SPARK-16043 will not be merged soon. This is because performance issues for DataFrame/Dataset programs with primitive arrays are addressed by other approaches.
If there are some bench programs for this measurement, I am happy to run them with SPARK-16043. Are there any benchmark programs?

> Performance of Parquet encodings on saving primitive arrays
> -----------------------------------------------------------
>
>                 Key: SPARK-16073
>                 URL: https://issues.apache.org/jira/browse/SPARK-16073
>             Project: Spark
>          Issue Type: Task
>          Components: MLlib, SQL
>    Affects Versions: 2.0.0
>            Reporter: Xiangrui Meng
>
> Spark supports both uncompressed and compressed (snappy, gzip, lzo) Parquet data. However, Parquet also has its own encodings to compress columns/arrays, e.g., dictionary encoding: https://github.com/apache/parquet-format/blob/master/Encodings.md.
> It might be worth checking the performance overhead of Parquet encodings on saving large primitive arrays, which is a machine learning use case. If the overhead is significant, we should expose a configuration in Spark to control the encoding levels.
> Note that this shouldn't be tested under Spark until SPARK-16043 was fixed.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org