[
https://issues.apache.org/jira/browse/HDFS-5027?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13719622#comment-13719622
]
Kihwal Lee commented on HDFS-5027:
----------------------------------
At this stage of datanode startup, it is mostly I/O bound. Even if the I/O
bandwidth is saturated by these threads, it won't slow down I/O at individual
disk level, which is the major factor of startup time in this phase. If these
were threads that are fighting for shared spindles, we might consider limiting
the number as we do for DataXceivers. But they are not.
> On startup, DN should scan volumes in parallel
> ----------------------------------------------
>
> Key: HDFS-5027
> URL: https://issues.apache.org/jira/browse/HDFS-5027
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: datanode
> Affects Versions: 2.0.4-alpha
> Reporter: Aaron T. Myers
> Assignee: Aaron T. Myers
> Attachments: HDFS-5027.patch
>
>
> On startup the DN must scan all replicas on all configured volumes before the
> initial block report to the NN. This is currently done serially, but can be
> done in parallel to improve startup time of the DN.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
