[
https://issues.apache.org/jira/browse/LUCENE-9613?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17404419#comment-17404419
]
Adrien Grand commented on LUCENE-9613:
--------------------------------------
I pushed some more specialization that gave the following results on
wikimedium10m.
{noformat}
TaskQPS baseline StdDev QPS patch StdDev
Pct diff p-value
HighSloppyPhrase 24.30 (6.9%) 23.50 (9.3%)
-3.3% ( -18% - 13%) 0.204
MedSloppyPhrase 70.81 (4.4%) 69.37 (5.4%)
-2.0% ( -11% - 8%) 0.191
LowSloppyPhrase 30.96 (3.3%) 30.49 (3.6%)
-1.5% ( -8% - 5%) 0.163
HighTermTitleBDVSort 19.49 (2.8%) 19.38 (2.6%)
-0.6% ( -5% - 4%) 0.503
MedPhrase 322.19 (3.8%) 321.87 (9.1%)
-0.1% ( -12% - 13%) 0.964
BrowseDayOfYearTaxoFacets 3.18 (3.7%) 3.18 (3.3%)
-0.0% ( -6% - 7%) 0.981
BrowseDateTaxoFacets 3.18 (3.7%) 3.18 (3.2%)
0.1% ( -6% - 7%) 0.952
BrowseMonthTaxoFacets 3.45 (4.7%) 3.46 (4.3%)
0.2% ( -8% - 9%) 0.895
IntNRQ 91.30 (43.5%) 91.51 (43.4%)
0.2% ( -60% - 154%) 0.987
LowIntervalsOrdered 17.60 (6.6%) 17.64 (7.1%)
0.2% ( -12% - 14%) 0.915
AndHighLow 1005.24 (4.1%) 1008.36 (4.0%)
0.3% ( -7% - 8%) 0.808
Prefix3 378.76 (11.9%) 380.28 (10.5%)
0.4% ( -19% - 25%) 0.910
LowPhrase 112.90 (2.8%) 113.37 (3.8%)
0.4% ( -6% - 7%) 0.694
HighSpanNear 51.40 (3.0%) 51.64 (3.0%)
0.5% ( -5% - 6%) 0.621
OrHighNotLow 1445.33 (4.9%) 1456.37 (4.6%)
0.8% ( -8% - 10%) 0.614
MedTerm 2527.24 (6.3%) 2548.62 (4.6%)
0.8% ( -9% - 12%) 0.628
OrNotHighMed 1157.13 (2.7%) 1167.00 (3.3%)
0.9% ( -5% - 7%) 0.370
LowSpanNear 44.09 (2.0%) 44.48 (2.1%)
0.9% ( -3% - 5%) 0.184
MedIntervalsOrdered 10.95 (3.4%) 11.04 (3.5%)
0.9% ( -5% - 8%) 0.420
HighIntervalsOrdered 25.53 (3.6%) 25.77 (4.1%)
1.0% ( -6% - 8%) 0.435
MedSpanNear 109.47 (2.0%) 110.57 (2.7%)
1.0% ( -3% - 5%) 0.183
OrHighNotHigh 1095.98 (4.0%) 1107.45 (3.4%)
1.0% ( -6% - 8%) 0.373
Fuzzy1 212.12 (6.8%) 214.37 (6.3%)
1.1% ( -11% - 15%) 0.609
OrHighHigh 34.88 (4.7%) 35.26 (3.2%)
1.1% ( -6% - 9%) 0.392
OrHighMed 124.51 (4.6%) 125.91 (2.5%)
1.1% ( -5% - 8%) 0.339
Respell 271.84 (3.0%) 274.94 (2.8%)
1.1% ( -4% - 7%) 0.210
OrHighNotMed 1397.92 (4.0%) 1414.46 (3.9%)
1.2% ( -6% - 9%) 0.344
HighPhrase 674.43 (2.1%) 682.48 (4.1%)
1.2% ( -4% - 7%) 0.245
AndHighHigh 53.28 (3.4%) 53.92 (4.0%)
1.2% ( -6% - 8%) 0.308
OrHighLow 477.86 (4.1%) 483.78 (3.5%)
1.2% ( -6% - 9%) 0.308
OrNotHighLow 1223.79 (3.8%) 1239.31 (4.3%)
1.3% ( -6% - 9%) 0.321
AndHighMed 106.80 (3.5%) 108.17 (3.9%)
1.3% ( -5% - 8%) 0.271
LowTerm 2514.56 (5.8%) 2549.48 (6.6%)
1.4% ( -10% - 14%) 0.478
Wildcard 157.42 (3.9%) 159.71 (4.1%)
1.5% ( -6% - 9%) 0.246
OrNotHighHigh 1013.51 (3.2%) 1028.66 (3.7%)
1.5% ( -5% - 8%) 0.176
Fuzzy2 154.94 (8.8%) 157.37 (8.4%)
1.6% ( -14% - 20%) 0.565
HighTerm 1590.75 (4.9%) 1624.88 (4.9%)
2.1% ( -7% - 12%) 0.168
HighTermMonthSort 78.11 (7.6%) 81.58 (9.1%)
4.4% ( -11% - 22%) 0.093
TermDTSort 84.05 (7.2%) 87.88 (7.1%)
4.5% ( -9% - 20%) 0.044
HighTermDayOfYearSort 116.77 (6.1%) 122.33 (6.8%)
4.8% ( -7% - 18%) 0.020
BrowseMonthSSDVFacets 12.98 (3.1%) 14.45 (5.1%)
11.3% ( 2% - 20%) 0.000
BrowseDayOfYearSSDVFacets 12.38 (3.5%) 15.52 (12.7%)
25.3% ( 8% - 43%) 0.000
{noformat}
> Create blocks for ords when it helps in Lucene80DocValuesFormat
> ---------------------------------------------------------------
>
> Key: LUCENE-9613
> URL: https://issues.apache.org/jira/browse/LUCENE-9613
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Priority: Minor
> Fix For: main (9.0)
>
> Time Spent: 1h 10m
> Remaining Estimate: 0h
>
> Currently for sorted(-set) values, we always write ords using
> log2(valueCount) bits per entry. However in several cases like when the field
> is used in the index sort, or if one value is _very_common, splitting into
> blocks like we do for numerics would help.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
