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-5554) Add TermBulkScorer
Date Thu, 27 Mar 2014 15:53:17 GMT

    [ https://issues.apache.org/jira/browse/LUCENE-5554?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13949468#comment-13949468
] 

Michael McCandless commented on LUCENE-5554:
--------------------------------------------

I ran the same "pre vs post" perf test as above, this time with my original patch:

{noformat}
                    Task    QPS base      StdDev    QPS comp      StdDev                Pct
diff
                  IntNRQ        2.81      (4.6%)        2.61      (5.2%)   -7.0% ( -16% -
   2%)
                 Prefix3       72.54      (2.7%)       68.45      (3.5%)   -5.6% ( -11% -
   0%)
                Wildcard       15.13      (3.7%)       14.47      (3.5%)   -4.3% ( -11% -
   3%)
              AndHighLow      296.42      (3.1%)      286.01      (3.5%)   -3.5% (  -9% -
   3%)
            HighSpanNear        6.14      (6.2%)        6.00      (6.0%)   -2.3% ( -13% -
  10%)
             AndHighHigh       23.06      (2.2%)       22.77      (2.4%)   -1.3% (  -5% -
   3%)
             MedSpanNear       24.03      (4.3%)       23.73      (4.1%)   -1.2% (  -9% -
   7%)
              AndHighMed       27.80      (2.6%)       27.51      (2.4%)   -1.1% (  -5% -
   4%)
        HighSloppyPhrase        2.63      (5.7%)        2.60      (7.0%)   -1.0% ( -12% -
  12%)
             LowSpanNear        8.17      (4.1%)        8.10      (4.1%)   -0.9% (  -8% -
   7%)
               OrHighLow       17.03      (2.9%)       16.90      (6.5%)   -0.7% (  -9% -
   8%)
            OrHighNotMed       26.17      (2.7%)       25.98      (6.2%)   -0.7% (  -9% -
   8%)
           OrHighNotHigh       10.36      (2.7%)       10.30      (6.9%)   -0.6% (  -9% -
   9%)
         MedSloppyPhrase        2.85      (5.3%)        2.83      (4.2%)   -0.5% (  -9% -
   9%)
         LowSloppyPhrase       35.49      (2.5%)       35.32      (2.1%)   -0.5% (  -4% -
   4%)
            OrNotHighLow       19.51      (3.0%)       19.43      (7.3%)   -0.4% ( -10% -
  10%)
            OrHighNotLow       22.14      (2.7%)       22.06      (6.2%)   -0.4% (  -9% -
   8%)
               OrHighMed       23.18      (2.5%)       23.10      (6.0%)   -0.3% (  -8% -
   8%)
            OrNotHighMed       18.66      (2.9%)       18.62      (7.1%)   -0.2% (  -9% -
  10%)
           OrNotHighHigh        8.35      (3.0%)        8.35      (6.9%)   -0.1% (  -9% -
  10%)
              OrHighHigh        7.54      (2.6%)        7.54      (6.9%)   -0.0% (  -9% -
   9%)
                 Respell       37.21      (3.9%)       37.26      (4.2%)    0.1% (  -7% -
   8%)
              HighPhrase        3.22      (7.7%)        3.23      (7.6%)    0.3% ( -13% -
  16%)
               LowPhrase       10.27      (3.1%)       10.32      (2.7%)    0.5% (  -5% -
   6%)
               MedPhrase      156.68      (7.1%)      158.03      (6.9%)    0.9% ( -12% -
  15%)
                  Fuzzy2       34.20      (3.2%)       34.80      (3.0%)    1.8% (  -4% -
   8%)
                  Fuzzy1       43.47      (3.3%)       44.63      (2.8%)    2.7% (  -3% -
   9%)
                 LowTerm      216.83     (15.8%)      227.80      (9.3%)    5.1% ( -17% -
  35%)
                 MedTerm       71.04     (11.6%)       77.63      (8.7%)    9.3% (  -9% -
  33%)
                HighTerm       49.15     (11.7%)       57.07     (12.5%)   16.1% (  -7% -
  45%)
{noformat}

Looks like similar gains as Rob's patch, except TermQuery saw a bigger speedup ...

Rob had a good idea: to try having IndexSearcher handle the default bulk-scoring case, instead
of Weight.bulkScorer creating a DefaultBulkScorer ... I'll try that next.

> Add TermBulkScorer
> ------------------
>
>                 Key: LUCENE-5554
>                 URL: https://issues.apache.org/jira/browse/LUCENE-5554
>             Project: Lucene - Core
>          Issue Type: Improvement
>          Components: core/search
>            Reporter: Michael McCandless
>            Assignee: Michael McCandless
>             Fix For: 4.8, 5.0
>
>         Attachments: LUCENE-5554.patch, LUCENE-5554.patch
>
>
> Hotspot was unhappy with the changes in LUCENE-5487, e.g.: http://people.apache.org/~mikemccand/lucenebench/OrHighHigh.html
> But it looks like we can get the performance back by making a dedicated BulkScorer for
TermQuery.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

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


Mime
View raw message