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-8695) TreeAggregation shouldn't be triggered when it doesn't save wall-clock time
Date Sun, 19 Jul 2015 09:59:04 GMT

    [ https://issues.apache.org/jira/browse/SPARK-8695?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14632760#comment-14632760
] 

Sean Owen commented on SPARK-8695:
----------------------------------

PS Xiangrui says: 

I think the right logic is numPartitions > scale + math.ceil(1.0 * numPartitions / scale).

> TreeAggregation shouldn't be triggered when it doesn't save wall-clock time
> ---------------------------------------------------------------------------
>
>                 Key: SPARK-8695
>                 URL: https://issues.apache.org/jira/browse/SPARK-8695
>             Project: Spark
>          Issue Type: Improvement
>          Components: MLlib, Spark Core
>    Affects Versions: 1.5.0
>            Reporter: Xiangrui Meng
>            Priority: Minor
>              Labels: starter
>
> If an RDD has 5 partitions, tree aggregation doesn't reduce the wall-clock time. Instead,
it introduces scheduling and shuffling overhead. We should update the condition to use tree
aggregation (code attached):
> {code}
>       while (numPartitions > scale + numPartitions / scale) {
> {code}



--
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