[
https://issues.apache.org/jira/browse/HDFS-13046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16357246#comment-16357246
]
Xiao Chen commented on HDFS-13046:
----------------------------------
+0 Mixed feelings but given this is off by default it should be fine.
As Akira said if network is more performant it would be a good idea to separate
the reads to more DN, so we get an overall higher utilization. (Best fit for
EC?)
On the other hand, it's also possible that we schedule more reads to be remote
and sacrifices locality, In the case when network is slow, the overall read
throughput may be lower.
> consider load of datanodes when read blocks of file
> ---------------------------------------------------
>
> Key: HDFS-13046
> URL: https://issues.apache.org/jira/browse/HDFS-13046
> Project: Hadoop HDFS
> Issue Type: New Feature
> Reporter: hu xiaodong
> Assignee: hu xiaodong
> Priority: Minor
> Attachments:
> HDFS-13046-considerLoadAfterSortBydistance-001-sample.patch,
> HDFS-13046-sample.patch
>
>
> When sorting block locations, we just consider the distance of datanodes. can
> we consider the load of datanodes? We can add a configuration such as
> 'dfs.namenode.reading.considerLoad', if set to true, then sort the
> blocklocations by load of the datanodes, otherwise sort by distance.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
