spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sean Owen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-1980) problems introduced by broadcast
Date Fri, 23 Jan 2015 12:53:34 GMT

    [ https://issues.apache.org/jira/browse/SPARK-1980?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14289210#comment-14289210
] 

Sean Owen commented on SPARK-1980:
----------------------------------

This error doesn't say anything except that the app died. What was the underlying problem
-- running out of memory? was that due to your application, configuration, or due to some
Spark problem? are you saying there is a problem cleaning up broadcasts?

> problems introduced by broadcast
> --------------------------------
>
>                 Key: SPARK-1980
>                 URL: https://issues.apache.org/jira/browse/SPARK-1980
>             Project: Spark
>          Issue Type: Bug
>            Reporter: zhoudi
>
> i am writing a word embedding on SPARK. The scale of the model is about 60w * 100  *
Float.size. Because of the large scale, I have to use broadcast to deliver the current model
to executors. After each iteration, I would update the model, and then broadcast it again.
The pseudo-code is as follows,
> for (i <- 0 to 100) {
>     broadcast_model <- broadcast(model);       
>     e_model = xxx.map(Func(broadcast_model)) // handle of broadcast_model to Func;  
    
>                           .reduce(_ + _)     
>     model <- model + e_model // Update the model
> }
> My problem is that an Error would come out after six iteration. The Error Info is as
follows,
> ./bin/spark-submit: line 44: 28232 killed                    $SPARK_HOME/bin/spark-class
org.apache.spark.deploy.Spark "${ORIG_ARGS[@]}"        



--
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


Mime
View raw message