[
https://issues.apache.org/jira/browse/HADOOP-10410?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13942185#comment-13942185
]
Andrew Wang commented on HADOOP-10410:
--------------------------------------
Hey Haohui, I can speak a little on this,
One nifty thing you can do with ioprio is set priorities on a per-thread basis.
This lets you do differentiated service among the DataXceivers in a datanode
based on (for example) a QoS field or a username.
Regarding capabilities, I think you only need CAP_SYS_ADMIN to set things in
the RT class. You can still use the normal priority class as long as you're the
owner of the process. The ioprio_set page talks about this
(http://man7.org/linux/man-pages/man2/ioprio_set.2.html)
I don't have my numbers from my ioprio prototype anymore, but I found pretty
nice gains in 99th percentile latency after I zeroed out the disk queue. I
didn't bother submitting it upstream at the time since I thought the throughput
hit wasn't worth it, but Liang said that latency is his foremost concern.
> Support ioprio_set in NativeIO
> ------------------------------
>
> Key: HADOOP-10410
> URL: https://issues.apache.org/jira/browse/HADOOP-10410
> Project: Hadoop Common
> Issue Type: New Feature
> Components: native
> Affects Versions: 3.0.0, 2.4.0
> Reporter: Liang Xie
> Assignee: Liang Xie
> Attachments: HADOOP-10410.txt
>
>
> It would be better to HBase application if HDFS layer provide a fine-grained
> IO request priority. Most of modern kernel should support ioprio_set system
> call now.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
