[
https://issues.apache.org/jira/browse/HDFS-5675?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13851576#comment-13851576
]
Steve Loughran commented on HDFS-5675:
--------------------------------------
# this is testing depth of directory creation too, so see if the create time is
O(1) vs O(depth) vs something bigger? As HDFS acquires one lock for the entire
operation, it should be O(1), but native filesystems usually have an O(depth)
action, as would blobstores.
# may be interesting to measure performance of mkdirs when the target directory
actually exists, to make sure that the cost of that operation doesn't increase
due to some regression (locking, presumably)
> Add Mkdirs operation to NNThroughputBenchmark
> ---------------------------------------------
>
> Key: HDFS-5675
> URL: https://issues.apache.org/jira/browse/HDFS-5675
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: benchmarks
> Reporter: Plamen Jeliazkov
> Assignee: Plamen Jeliazkov
> Priority: Minor
> Attachments: mkdirsBenchmarkPatchTrunk.patch,
> mkdirsBenchmarkPatchTrunk_2.patch
>
>
> I did some work to extend NNThroughputBenchmark that I would like to
> contribute to the community. It is pretty straightforward; just adding a
> Mkdir operation to the test in order to see the operations per second of a
> multiple 'mkdir' commands.
--
This message was sent by Atlassian JIRA
(v6.1.4#6159)
