[
https://issues.apache.org/jira/browse/HDFS-14731?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17042332#comment-17042332
]
Konstantin Shvachko commented on HDFS-14731:
--------------------------------------------
I also ran some NNThroughputBenchmarks following [~vagarychen]'s suggestion to
evaluate NN performance with this patch.
# Main thing is that multi-threading testing didn't show any errors.
# On write operations there was no visible difference between current trunk and
with my patch:
mkdir, delete QPS were around 26K ops/sec, while create was 20K for both
versions
# With read operations fileStatus and getBlockLocations I see consistently
better QPS with the patch.
On trunk QPS varies between 80K-130K. While with the patch it is in the range
120K-170K.
So I would propose to commit this to earlier branches as well. What people
think?
> [FGL] Remove redundant locking on NameNode.
> -------------------------------------------
>
> Key: HDFS-14731
> URL: https://issues.apache.org/jira/browse/HDFS-14731
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: namenode
> Reporter: Konstantin Shvachko
> Assignee: Konstantin Shvachko
> Priority: Major
> Attachments: HDFS-14731.001.patch
>
>
> Currently NameNode has two global locks: FSNamesystemLock and
> FSDirectoryLock. An analysis shows that single FSNamesystemLock is sufficient
> to guarantee consistency of the NameNode state. FSDirectoryLock can be
> removed.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
