[
https://issues.apache.org/jira/browse/LUCENE-4864?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13611369#comment-13611369
]
Michael Poindexter commented on LUCENE-4864:
--------------------------------------------
I really think it would be worth running a test on windows with a larger number
of threads (3 seems quite low for something like a web app where there could be
a large number of concurrent users). The implementation on Windows in the JRE
is very, very different internally, so I'm not sure I'm quite as sure as you
are that this would necessarily be the same.
I'd love to run some tests on windows with different concurrency levels and
play around with tweaking the executor service used for this to see if it can
have better performance than SimpleFSDirectory/NIOFSDirectory if I had a way to
do that. If you guys think it's not worth it I'll drop it, but I'm willing to
put in the effort to get a test setup going and tweak things and report back
numbers if there is any interest and someone can point me in the right
direction to run a perf test.
If there is any interest, I'm going out of town for the next week, but I'll
check in on this when I get back and look at doing the tests if you guys want.
Thanks for the attention on this either way!
> Add AsyncFSDirectory to work around Windows issues with NIOFS (Lucene 5.0
> only)
> -------------------------------------------------------------------------------
>
> Key: LUCENE-4864
> URL: https://issues.apache.org/jira/browse/LUCENE-4864
> Project: Lucene - Core
> Issue Type: Improvement
> Components: core/store
> Affects Versions: 5.0
> Reporter: Michael Poindexter
> Attachments: LUCENE-4864.patch, LUCENE-4864.patch
>
>
> On LUCENE-4848 a new directory implementation was proposed that uses
> AsyncFileChannel to make a sync-less directory implementation (only needed
> for IndexInput). The problem on Windows is that positional reads are
> impossible without overlapping (async) I/O, so FileChannel in the JDK has to
> syncronize all reads, because they consist of an atomic seek and atomic read.
> AsyncFSDirectoty would not have this issue, but has to take care of thread
> management, because you need a separate thread to get notified when the read
> is done. This involves overhead, but might still be better than the
> synchronization.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
