mikemccand commented on issue #12527: URL: https://github.com/apache/lucene/issues/12527#issuecomment-1700871714
I like this idea, reducing possible IO overhead. But I tested it with
`luceneutil` on `wikimediumall`:
```
Task QPS base StdDev QPS base2
StdDev Pct diff p-value
IntNRQ 76.78 (10.7%) 62.62
(7.6%) -18.4% ( -33% - 0%) 0.000
OrHighLow 304.44 (4.9%) 298.02
(2.5%) -2.1% ( -9% - 5%) 0.085
OrHighHigh 34.42 (6.2%) 33.75
(4.6%) -1.9% ( -12% - 9%) 0.265
OrHighMed 132.82 (4.7%) 130.46
(2.9%) -1.8% ( -9% - 6%) 0.153
HighTerm 774.84 (5.6%) 764.08
(5.7%) -1.4% ( -12% - 10%) 0.437
MedTerm 903.16 (5.6%) 890.65
(5.7%) -1.4% ( -12% - 10%) 0.438
OrHighNotLow 513.31 (7.3%) 506.82
(5.9%) -1.3% ( -13% - 12%) 0.546
BrowseRandomLabelSSDVFacets 9.36 (4.6%) 9.25
(5.0%) -1.2% ( -10% - 8%) 0.420
AndHighHigh 36.69 (3.1%) 36.28
(3.9%) -1.1% ( -7% - 6%) 0.311
AndHighMed 247.06 (2.4%) 244.49
(3.2%) -1.0% ( -6% - 4%) 0.251
OrNotHighHigh 446.63 (4.6%) 442.08
(3.8%) -1.0% ( -9% - 7%) 0.448
OrHighNotMed 506.58 (7.3%) 501.84
(6.4%) -0.9% ( -13% - 13%) 0.666
BrowseDateTaxoFacets 8.32 (5.8%) 8.25
(5.0%) -0.9% ( -10% - 10%) 0.611
BrowseDayOfYearTaxoFacets 8.29 (5.7%) 8.22
(4.9%) -0.9% ( -10% - 10%) 0.614
OrHighNotHigh 457.35 (5.8%) 453.90
(4.9%) -0.8% ( -10% - 10%) 0.658
OrNotHighMed 412.12 (2.5%) 409.12
(2.0%) -0.7% ( -5% - 3%) 0.315
LowTerm 705.46 (3.6%) 701.45
(3.9%) -0.6% ( -7% - 7%) 0.632
OrHighMedDayTaxoFacets 9.38 (3.5%) 9.34
(4.6%) -0.5% ( -8% - 7%) 0.715
HighSloppyPhrase 10.43 (3.4%) 10.39
(3.4%) -0.4% ( -6% - 6%) 0.714
BrowseRandomLabelTaxoFacets 7.59 (3.8%) 7.56
(3.3%) -0.4% ( -7% - 6%) 0.731
MedSpanNear 78.40 (2.6%) 78.11
(1.8%) -0.4% ( -4% - 4%) 0.589
BrowseDateSSDVFacets 2.18 (1.6%) 2.18
(1.1%) -0.4% ( -3% - 2%) 0.399
HighTermTitleBDVSort 9.16 (1.7%) 9.13
(1.9%) -0.3% ( -3% - 3%) 0.551
PKLookup 253.13 (0.9%) 252.32
(0.8%) -0.3% ( -1% - 1%) 0.215
OrNotHighLow 428.27 (2.2%) 427.00
(1.6%) -0.3% ( -4% - 3%) 0.627
LowSpanNear 16.45 (2.5%) 16.41
(1.7%) -0.3% ( -4% - 4%) 0.695
HighIntervalsOrdered 12.58 (2.9%) 12.55
(3.5%) -0.3% ( -6% - 6%) 0.797
HighSpanNear 19.81 (2.1%) 19.78
(1.7%) -0.2% ( -3% - 3%) 0.794
LowSloppyPhrase 38.11 (1.9%) 38.06
(2.1%) -0.1% ( -4% - 3%) 0.846
HighTermTitleSort 263.09 (2.4%) 262.91
(2.2%) -0.1% ( -4% - 4%) 0.927
TermDTSort 482.30 (0.8%) 482.18
(0.7%) -0.0% ( -1% - 1%) 0.919
HighTermMonthSort 3449.43 (1.1%) 3448.77
(0.7%) -0.0% ( -1% - 1%) 0.948
AndHighLow 863.72 (0.9%) 863.69
(1.2%) -0.0% ( -2% - 2%) 0.991
MedSloppyPhrase 46.06 (4.3%) 46.06
(4.4%) -0.0% ( -8% - 9%) 1.000
MedPhrase 32.96 (2.6%) 32.96
(2.9%) 0.0% ( -5% - 5%) 1.000
Fuzzy2 53.68 (1.2%) 53.68
(1.2%) 0.0% ( -2% - 2%) 0.995
BrowseMonthTaxoFacets 8.25 (0.2%) 8.25
(0.2%) 0.0% ( 0% - 0%) 0.789
HighTermDayOfYearSort 751.68 (0.8%) 752.01
(0.8%) 0.0% ( -1% - 1%) 0.866
AndHighHighDayTaxoFacets 7.74 (2.5%) 7.75
(1.7%) 0.1% ( -4% - 4%) 0.935
LowIntervalsOrdered 26.92 (2.0%) 26.94
(2.2%) 0.1% ( -4% - 4%) 0.900
MedTermDayTaxoFacets 31.71 (2.9%) 31.74
(2.3%) 0.1% ( -4% - 5%) 0.912
Wildcard 38.55 (2.2%) 38.60
(1.9%) 0.1% ( -3% - 4%) 0.843
MedIntervalsOrdered 4.74 (2.2%) 4.74
(2.2%) 0.2% ( -4% - 4%) 0.817
BrowseMonthSSDVFacets 14.04 (9.1%) 14.07
(11.7%) 0.2% ( -18% - 23%) 0.948
Respell 39.73 (1.2%) 39.81
(1.3%) 0.2% ( -2% - 2%) 0.568
Fuzzy1 62.49 (1.3%) 62.67
(1.4%) 0.3% ( -2% - 2%) 0.484
AndHighMedDayTaxoFacets 42.88 (1.4%) 43.02
(1.1%) 0.3% ( -2% - 2%) 0.431
LowPhrase 20.89 (3.3%) 20.95
(3.8%) 0.3% ( -6% - 7%) 0.775
HighPhrase 66.24 (3.4%) 66.51
(3.6%) 0.4% ( -6% - 7%) 0.713
Prefix3 198.81 (2.2%) 199.70
(1.8%) 0.4% ( -3% - 4%) 0.487
BrowseDayOfYearSSDVFacets 12.50 (4.8%) 12.73
(10.6%) 1.8% ( -12% - 18%) 0.487
```
It looks like `IntNRQ` (which I think is the only tasks using BKD tree /
points for range filtering) is upset with it with high confidence (`p=0.000`).
I'm surprised the impact was THAT large.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
