lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Michael McCandless (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (LUCENE-3023) Land DWPT on trunk
Date Tue, 03 May 2011 17:25:04 GMT

    [ https://issues.apache.org/jira/browse/LUCENE-3023?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13028327#comment-13028327
] 

Michael McCandless commented on LUCENE-3023:
--------------------------------------------

This cutover to concurrent flushing (DWPT) produces astounding
increases in indexing throughput:

  http://people.apache.org/~mikemccand/lucenebench/indexing.html

186 GB plain text per hour (from 101 GB/hour the day before)!!!

It's not every day you see an 84% jump in indexing throughput!  Wow.

This is on a machine that has substantial CPU+IO concurrency, ie, it
was bottlenecked by our non-concurrent flush.

Also, I can now tune up the IW settings I use in those nightly
benchmarks; it's now 6 threads and "only" 512 MB RAM buffer.  I'll
wait a few days and then do that.

Looks like a few queries got a bit slower... I suspect this is because
the index segment count has changed.  Before concurrent flushing it
was this:
{noformat}
36(4.0):C4977400 _69(4.0):C4977400 _9c(4.0):C4977400 _cf(4.0):C4977400 _fi(4.0):C4977400 _fq(4.0):C497740
_g1(4.0):C497740 _gc(4.0):C497740 _gn(4.0):C497740 _gy(4.0):C497740 _gx(4.0):C49774 _gz(4.0):C49774
_h0(4.0):C49774 _h1(4.0):C49774 _h2(4.0):C49774 _h3(4.0):C468
{noformat}

After concurrent flushing:
{noformat}
_3d(4.0):C4977400 _6h(4.0):C4977400 _9j(4.0):C4977400 _cn(4.0):C4977400 _fq(4.0):C4977400
_fu(4.0):C497740 _g6(4.0):C497740 _gh(4.0):C497740 _gs(4.0):C497740 _h2(4.0):C497740 _gy(4.0):C49774
_gz(4.0):C49774 _h0(4.0):C49774 _h5(4.0):C4105 _1(4.0):C2627 _h4(4.0):C16331 _h3(4.0):C28728
_h1(4.0):C48225
{noformat}

So we have 2 extra segments... it's interesting how this affects some
queries but not others.


> Land DWPT on trunk
> ------------------
>
>                 Key: LUCENE-3023
>                 URL: https://issues.apache.org/jira/browse/LUCENE-3023
>             Project: Lucene - Java
>          Issue Type: Task
>    Affects Versions: CSF branch, 4.0
>            Reporter: Simon Willnauer
>            Assignee: Simon Willnauer
>             Fix For: 4.0
>
>         Attachments: LUCENE-3023-quicksort-reincarnation.patch, LUCENE-3023-svn-diff.patch,
LUCENE-3023-ws-changes.patch, LUCENE-3023.patch, LUCENE-3023.patch, LUCENE-3023.patch, LUCENE-3023.patch,
LUCENE-3023_CHANGES.patch, LUCENE-3023_CHANGES.patch, LUCENE-3023_iw_iwc_jdoc.patch, LUCENE-3023_simonw_review.patch,
LUCENE-3023_svndiff.patch, LUCENE-3023_svndiff.patch, diffMccand.py, diffSources.patch, diffSources.patch,
realtime-TestAddIndexes-3.txt, realtime-TestAddIndexes-5.txt, realtime-TestIndexWriterExceptions-assert-6.txt,
realtime-TestIndexWriterExceptions-npe-1.txt, realtime-TestIndexWriterExceptions-npe-2.txt,
realtime-TestIndexWriterExceptions-npe-4.txt, realtime-TestOmitTf-corrupt-0.txt
>
>
> With LUCENE-2956 we have resolved the last remaining issue for LUCENE-2324 so we can
proceed landing the DWPT development on trunk soon. I think one of the bigger issues here
is to make sure that all JavaDocs for IW etc. are still correct though. I will start going
through that first.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message