[
https://issues.apache.org/jira/browse/HDFS-5239?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15536471#comment-15536471
]
Zhe Zhang commented on HDFS-5239:
---------------------------------
[~daryn] While working on HDFS-10896 we are revisiting this config key
{{dfs.namenode.fslock.fair}}. Do you think we should properly expose and
document it as a config key now? [~xkrogen] Already included a draft of it in
one of his patches. Let us know how you think about it. For now we are going to
separate it out from the HDFS-10896 patch.
{code}
+<property>
+ <name>dfs.namenode.fslock.fair</name>
+ <value>true</value>
+ <description>If this is true, the FS Namesystem lock will be used in Fair
+ mode, which will help to prevent writer threads from being starved, but
+ can provide lower lock throughput.
+ </description>
+</property>
{code}
> Allow FSNamesystem lock fairness to be configurable
> ---------------------------------------------------
>
> Key: HDFS-5239
> URL: https://issues.apache.org/jira/browse/HDFS-5239
> Project: Hadoop HDFS
> Issue Type: Sub-task
> Components: namenode
> Affects Versions: 2.0.0-alpha, 3.0.0-alpha1
> Reporter: Daryn Sharp
> Assignee: Daryn Sharp
> Fix For: 0.23.10, 2.3.0
>
> Attachments: HADOOP-5239.branach-23.patch, HDFS-5239.patch
>
>
> The fairness of the {{FSNamesystem#fsLock}} is hardcoded to {{true}}. Using
> an unfair lock provides a negligible increase to throughput. However this is
> due to bottlenecks elsewhere in the system. In combination with other
> changes, such as RPC layer and audit logging, preliminary tests show up to a
> 5X improvement for a read heavy workload.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
