[
https://issues.apache.org/jira/browse/HDFS-1127?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12864177#action_12864177
]
dhruba borthakur commented on HDFS-1127:
----------------------------------------
I think Hairong mentions that HADOOP-6640 already has a fix for this one.
> Improve the speed of FileSystem object creation
> -----------------------------------------------
>
> Key: HDFS-1127
> URL: https://issues.apache.org/jira/browse/HDFS-1127
> Project: Hadoop HDFS
> Issue Type: Improvement
> Affects Versions: 0.20.1, 0.20.2
> Reporter: Zheng Shao
> Assignee: Zheng Shao
> Attachments: HDFS-1127.1.patch, HDFS-1127.branch-20.2.patch,
> HDFS-1127.branch-20.3.patch
>
>
> It takes a long time to create a lot of {{DistributedFileSystem}} object via
> {{FileSystem.create}} or {{FileSystem.createNewInstance}} function.
> In our experience with scribe-hdfs, it can take up to 2 minutes for 100 such
> calls from each node of a cluster of 40 nodes.
> The reason for the delay is mainly because of the lock on the CACHE object.
> We should release the lock while creating new FileSystem objects.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
