[jira] [Commented] (SPARK-13830) Fetch large directly result from executor is very slow

["Apache Spark (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SPARK-13830?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15191511#comment-15191511 ] 

Apache Spark commented on SPARK-13830:
--------------------------------------

User 'davies' has created a pull request for this issue:
https://github.com/apache/spark/pull/11659

> Fetch large directly result from executor is very slow
> ------------------------------------------------------
>
>                 Key: SPARK-13830
>                 URL: https://issues.apache.org/jira/browse/SPARK-13830
>             Project: Spark
>          Issue Type: Task
>          Components: Spark Core
>            Reporter: Davies Liu
>
> Given two task with 100+M result on each, it take more than 50 seconds to fetch the results.
> The RPC may be not designed to handle large block, we should use block manager for that. But currently this is based on spark.rpc.message.maxSize, which is usually very large (> 128M) for safe, it's too large for handling results.
> We also counting the time to fetch the direct result (also deserialize it) as schedule delay, it also make sense to only fetch much smaller blocks via DirectResult.



--
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