[
https://issues.apache.org/jira/browse/HDFS-8865?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14705828#comment-14705828
]
Kihwal Lee commented on HDFS-8865:
----------------------------------
Thanks for the review [~xyao].
1. Numbers for small/medium size name space: I do not have numbers for them,
but I expect them to be better than before. The test cases show it taking a
couple of milliseconds for a very small name space, so the overhead of
fork-join doesn't seem to be of too much concern. For very small namespaces,
the ones that take far less than a second to initialize, the multithreaded
initialization cpuld take a bit longer.
2. This involves conversion to slf4j. I started down this path, but ended up
needing to update {{IOUtils}} as well. Since it is not a matter of updating
just {{FSImage}}, I won't do it in this jira. Instead I will simply wrap it
with {{isDebugEnabled()}} check for now.
> Improve quota initialization performance
> ----------------------------------------
>
> Key: HDFS-8865
> URL: https://issues.apache.org/jira/browse/HDFS-8865
> Project: Hadoop HDFS
> Issue Type: Improvement
> Reporter: Kihwal Lee
> Assignee: Kihwal Lee
> Attachments: HDFS-8865.patch, HDFS-8865.v2.checkstyle.patch,
> HDFS-8865.v2.patch
>
>
> After replaying edits, the whole file system tree is recursively scanned in
> order to initialize the quota. For big name space, this can take a very long
> time. Since this is done during namenode failover, it also affects failover
> latency.
> By using the Fork-Join framework, I was able to greatly reduce the
> initialization time. The following is the test result using the fsimage from
> one of the big name nodes we have.
> || threads || seconds||
> | 1 (existing) | 55|
> | 1 (fork-join) | 68 |
> | 4 | 16 |
> | 8 | 8 |
> | 12 | 6 |
> | 16 | 5 |
> | 20 | 4 |
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
