You are viewing a plain text version of this content. The canonical link for it is here.

Posted to hdfs-issues@hadoop.apache.org by "Haohui Mai (JIRA)" <ji...@apache.org> on 2015/04/01 00:04:53 UTC

[jira] [Commented] (HDFS-8009) Signal congestion on the DataNode

    [ https://issues.apache.org/jira/browse/HDFS-8009?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14389494#comment-14389494 ] 

Haohui Mai commented on HDFS-8009:
----------------------------------

The v0 patch allows the datanode to signal congestion when the load of the system is over 1.5 times of the number of available cores, which means that there are at least 50% more processes are waiting to be scheduled in the OS. We found it an effective metric to signal that the DNs are undergoing massive amount of writes.

> Signal congestion on the DataNode
> ---------------------------------
>
>                 Key: HDFS-8009
>                 URL: https://issues.apache.org/jira/browse/HDFS-8009
>             Project: Hadoop HDFS
>          Issue Type: New Feature
>            Reporter: Haohui Mai
>            Assignee: Haohui Mai
>         Attachments: HDFS-8009.000.patch
>
>
> The DataNode should signal congestion (i.e. "I'm too busy") in the PipelineAck using the mechanism introduced in HDFS-7270.



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