You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Michael McCandless (JIRA)" <ji...@apache.org> on 2014/07/10 00:19:05 UTC
[jira] [Commented] (LUCENE-5808) clean up postingsreader
[ https://issues.apache.org/jira/browse/LUCENE-5808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14056846#comment-14056846 ]
Michael McCandless commented on LUCENE-5808:
--------------------------------------------
I really like these simplifications.
I ran perf test vs trunk:
{noformat}
Report after iter 19:
Task QPS base StdDev QPS comp StdDev Pct diff
OrHighHigh 9.74 (13.5%) 8.28 (18.8%) -15.0% ( -41% - 20%)
OrNotHighLow 23.71 (13.3%) 21.22 (16.1%) -10.5% ( -35% - 21%)
OrHighMed 31.79 (7.6%) 28.73 (14.9%) -9.6% ( -29% - 13%)
OrHighNotHigh 13.18 (12.8%) 12.16 (14.1%) -7.7% ( -30% - 21%)
HighTerm 67.90 (14.5%) 62.68 (19.0%) -7.7% ( -35% - 30%)
OrHighNotLow 29.26 (14.3%) 27.10 (14.2%) -7.4% ( -31% - 24%)
OrNotHighMed 22.57 (15.0%) 20.93 (15.5%) -7.3% ( -32% - 27%)
Prefix3 86.86 (10.2%) 81.99 (14.8%) -5.6% ( -27% - 21%)
OrNotHighHigh 10.41 (13.2%) 9.87 (14.7%) -5.2% ( -29% - 26%)
Fuzzy1 55.92 (10.0%) 53.24 (13.4%) -4.8% ( -25% - 20%)
HighSloppyPhrase 3.42 (14.6%) 3.26 (17.5%) -4.6% ( -31% - 32%)
HighSpanNear 9.37 (15.8%) 9.09 (19.9%) -3.0% ( -33% - 38%)
HighPhrase 4.33 (10.8%) 4.20 (16.8%) -2.9% ( -27% - 27%)
OrHighLow 21.82 (15.5%) 21.38 (13.9%) -2.1% ( -27% - 32%)
AndHighMed 34.04 (4.8%) 33.56 (11.8%) -1.4% ( -17% - 15%)
OrHighNotMed 33.92 (19.4%) 33.57 (13.1%) -1.0% ( -28% - 38%)
LowTerm 318.33 (15.4%) 318.45 (12.9%) 0.0% ( -24% - 33%)
Respell 45.80 (11.8%) 45.85 (14.5%) 0.1% ( -23% - 29%)
AndHighHigh 28.10 (6.3%) 28.19 (11.3%) 0.3% ( -16% - 19%)
Fuzzy2 41.95 (10.1%) 42.40 (16.0%) 1.1% ( -22% - 30%)
Wildcard 18.84 (11.5%) 19.13 (12.2%) 1.5% ( -19% - 28%)
IntNRQ 3.17 (14.0%) 3.22 (17.2%) 1.5% ( -26% - 38%)
LowPhrase 12.83 (10.3%) 13.08 (16.3%) 2.0% ( -22% - 31%)
MedTerm 98.48 (18.2%) 100.63 (16.9%) 2.2% ( -27% - 45%)
MedPhrase 197.18 (13.3%) 201.95 (12.6%) 2.4% ( -20% - 32%)
MedSloppyPhrase 3.32 (16.1%) 3.50 (14.2%) 5.4% ( -21% - 42%)
AndHighLow 352.18 (12.9%) 375.73 (14.4%) 6.7% ( -18% - 39%)
LowSloppyPhrase 42.72 (12.0%) 46.52 (21.5%) 8.9% ( -22% - 48%)
LowSpanNear 10.23 (15.7%) 11.24 (18.9%) 9.9% ( -21% - 52%)
MedSpanNear 31.53 (14.1%) 35.40 (20.7%) 12.3% ( -19% - 54%)
{noformat}
Seems like OR queries lost a bit ...
> clean up postingsreader
> -----------------------
>
> Key: LUCENE-5808
> URL: https://issues.apache.org/jira/browse/LUCENE-5808
> Project: Lucene - Core
> Issue Type: Task
> Reporter: Robert Muir
> Attachments: LUCENE-5808.patch
>
>
> The current postingsreader is ~ 1500 lines of code (mostly duplicated) calling something like 4,000 lines of generated decompression code.
> This is really heavyweight and complicated, and bloats the lucene jar. It would be nice to simplify it so we can eventually remove the baggage.
--
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