[
https://issues.apache.org/jira/browse/LUCENE-5425?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13891972#comment-13891972
]
Michael McCandless commented on LUCENE-5425:
--------------------------------------------
OK I re-ran with the last patch:
{noformat}
Task QPS base StdDev QPS comp StdDev
Pct diff
HighPhrase 3.64 (5.0%) 3.56 (5.8%)
-2.2% ( -12% - 9%)
MedPhrase 145.39 (4.2%) 142.65 (4.6%)
-1.9% ( -10% - 7%)
OrNotHighLow 23.53 (5.4%) 23.30 (6.5%)
-1.0% ( -12% - 11%)
AndHighLow 290.69 (1.9%) 289.32 (1.6%)
-0.5% ( -3% - 3%)
OrNotHighMed 18.22 (4.5%) 18.16 (5.3%)
-0.3% ( -9% - 9%)
LowPhrase 12.03 (2.0%) 12.02 (1.9%)
-0.1% ( -3% - 3%)
OrNotHighHigh 5.88 (4.0%) 5.87 (4.3%)
-0.1% ( -8% - 8%)
OrHighHigh 4.84 (3.8%) 4.84 (4.0%)
-0.1% ( -7% - 8%)
AndHighMed 29.55 (1.8%) 29.53 (1.7%)
-0.1% ( -3% - 3%)
AndHighHigh 24.03 (1.6%) 24.03 (1.5%)
-0.0% ( -3% - 3%)
OrHighNotHigh 7.22 (3.7%) 7.22 (4.0%)
0.0% ( -7% - 7%)
OrHighLow 10.01 (3.6%) 10.04 (3.8%)
0.2% ( -6% - 7%)
LowSloppyPhrase 37.69 (1.8%) 37.78 (1.8%)
0.3% ( -3% - 3%)
LowSpanNear 9.56 (3.2%) 9.60 (3.6%)
0.4% ( -6% - 7%)
LowTerm 84.44 (3.5%) 84.82 (3.4%)
0.5% ( -6% - 7%)
MedSpanNear 26.26 (3.0%) 26.38 (3.0%)
0.5% ( -5% - 6%)
HighSpanNear 6.49 (5.0%) 6.52 (4.5%)
0.5% ( -8% - 10%)
Fuzzy1 64.43 (2.1%) 64.78 (1.6%)
0.5% ( -3% - 4%)
OrHighNotLow 12.56 (3.4%) 12.63 (3.5%)
0.5% ( -6% - 7%)
Fuzzy2 44.54 (2.2%) 44.80 (2.0%)
0.6% ( -3% - 4%)
OrHighMed 13.49 (3.2%) 13.57 (3.6%)
0.6% ( -5% - 7%)
Respell 47.31 (3.1%) 47.60 (2.5%)
0.6% ( -4% - 6%)
MedSloppyPhrase 3.26 (4.3%) 3.28 (5.0%)
0.7% ( -8% - 10%)
OrHighNotMed 15.62 (2.9%) 15.78 (3.4%)
1.0% ( -5% - 7%)
HighSloppyPhrase 3.15 (7.9%) 3.18 (11.1%)
1.0% ( -16% - 21%)
HighTerm 21.55 (2.6%) 21.79 (2.6%)
1.1% ( -3% - 6%)
MedTerm 31.23 (2.3%) 31.67 (2.6%)
1.4% ( -3% - 6%)
IntNRQ 4.11 (2.8%) 4.30 (2.3%)
4.7% ( 0% - 9%)
Wildcard 17.39 (2.6%) 20.99 (3.9%)
20.7% ( 13% - 27%)
Prefix3 75.17 (1.9%) 96.43 (3.3%)
28.3% ( 22% - 34%)
{noformat}
I think the improved FixedBitSet iterator made a huge difference for the MTQs
(wildcard, prefix), unrelated to facets I think, since those queries very
likely used the filter rewrite method.
And then there's no more ~10% hit for fast queries, which is great: DISI is
fine to use.
> Make creation of FixedBitSet in FacetsCollector overridable
> -----------------------------------------------------------
>
> Key: LUCENE-5425
> URL: https://issues.apache.org/jira/browse/LUCENE-5425
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/facet
> Affects Versions: 4.6
> Reporter: John Wang
> Attachments: LUCENE-5425.patch, facetscollector.patch,
> facetscollector.patch, fixbitset.patch
>
>
> In FacetsCollector, creation of bits in MatchingDocs are allocated per query.
> For large indexes where maxDocs are large creating a bitset of maxDoc bits
> will be expensive and would great a lot of garbage.
> Attached patch is to allow for this allocation customizable while maintaining
> current behavior.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
