iverase commented on pull request #2094:
URL: https://github.com/apache/lucene-solr/pull/2094#issuecomment-741872764
I changed PackedInts so they work on Little endian but still have the
capability of reading data stored in big endian. This makes the PR even bigger
but now we can run Lucene performance test directly. All test pass.
I have run luceneutil wikimediall test and here are the results:
```
TaskQPS baseline StdDevQPS my_modified_version
StdDev Pct diff p-value
Fuzzy2 71.59 (8.0%) 70.39 (7.9%)
-1.7% ( -16% - 15%) 0.504
Fuzzy1 62.58 (6.0%) 62.10 (8.0%)
-0.8% ( -13% - 14%) 0.731
BrowseMonthTaxoFacets 1.07 (3.8%) 1.06 (2.5%)
-0.7% ( -6% - 5%) 0.503
BrowseDateTaxoFacets 1.01 (3.7%) 1.01 (2.2%)
-0.5% ( -6% - 5%) 0.631
PKLookup 200.93 (4.3%) 200.15 (3.7%)
-0.4% ( -8% - 7%) 0.759
BrowseDayOfYearTaxoFacets 1.01 (3.7%) 1.00 (2.2%)
-0.4% ( -6% - 5%) 0.690
HighSloppyPhrase 2.54 (4.2%) 2.53 (4.2%)
-0.3% ( -8% - 8%) 0.833
LowSpanNear 20.41 (3.2%) 20.37 (3.7%)
-0.2% ( -6% - 6%) 0.846
Respell 42.35 (1.9%) 42.26 (1.6%)
-0.2% ( -3% - 3%) 0.702
HighIntervalsOrdered 1.10 (1.6%) 1.10 (1.6%)
-0.2% ( -3% - 3%) 0.745
OrHighNotMed 740.11 (3.5%) 738.98 (2.8%)
-0.2% ( -6% - 6%) 0.880
Prefix3 81.76 (5.3%) 81.71 (5.0%)
-0.1% ( -9% - 10%) 0.969
HighSpanNear 13.40 (2.4%) 13.40 (2.8%)
0.0% ( -5% - 5%) 0.954
MedTerm 1483.15 (3.3%) 1484.29 (4.1%)
0.1% ( -7% - 7%) 0.947
OrHighNotHigh 576.35 (4.8%) 577.11 (3.1%)
0.1% ( -7% - 8%) 0.917
AndHighHigh 39.42 (1.8%) 39.48 (2.2%)
0.2% ( -3% - 4%) 0.805
HighTermDayOfYearSort 68.23 (10.3%) 68.35 (10.2%)
0.2% ( -18% - 23%) 0.957
MedSpanNear 77.71 (2.4%) 77.85 (2.6%)
0.2% ( -4% - 5%) 0.817
MedSloppyPhrase 38.01 (2.1%) 38.09 (2.3%)
0.2% ( -4% - 4%) 0.754
LowTerm 1496.23 (2.8%) 1499.53 (3.6%)
0.2% ( -6% - 6%) 0.830
Wildcard 68.16 (8.0%) 68.32 (8.2%)
0.2% ( -14% - 17%) 0.930
AndHighMed 82.66 (2.6%) 82.96 (3.4%)
0.4% ( -5% - 6%) 0.707
BrowseMonthSSDVFacets 4.15 (5.0%) 4.16 (5.2%)
0.4% ( -9% - 11%) 0.807
MedPhrase 17.85 (2.0%) 17.92 (2.0%)
0.4% ( -3% - 4%) 0.519
OrNotHighHigh 638.37 (4.4%) 641.01 (3.8%)
0.4% ( -7% - 9%) 0.751
OrNotHighMed 718.57 (5.1%) 721.55 (3.4%)
0.4% ( -7% - 9%) 0.764
LowSloppyPhrase 21.19 (1.9%) 21.30 (2.1%)
0.5% ( -3% - 4%) 0.396
OrHighHigh 46.81 (1.6%) 47.06 (1.9%)
0.5% ( -2% - 4%) 0.343
OrHighNotLow 776.11 (4.8%) 780.40 (3.4%)
0.6% ( -7% - 9%) 0.672
OrHighMed 38.36 (2.4%) 38.61 (2.9%)
0.7% ( -4% - 6%) 0.429
HighPhrase 423.52 (1.8%) 427.22 (1.9%)
0.9% ( -2% - 4%) 0.132
LowPhrase 56.12 (1.7%) 56.63 (2.1%)
0.9% ( -2% - 4%) 0.141
OrHighLow 326.11 (4.6%) 329.15 (6.4%)
0.9% ( -9% - 12%) 0.598
OrNotHighLow 457.65 (3.3%) 462.47 (4.2%)
1.1% ( -6% - 8%) 0.376
HighTerm 930.64 (3.7%) 941.65 (4.2%)
1.2% ( -6% - 9%) 0.342
HighTermMonthSort 41.85 (12.8%) 42.40 (12.5%)
1.3% ( -21% - 30%) 0.746
AndHighLow 461.07 (4.3%) 470.01 (5.8%)
1.9% ( -7% - 12%) 0.231
HighTermTitleBDVSort 38.12 (12.0%) 38.93 (16.5%)
2.1% ( -23% - 34%) 0.641
IntNRQ 81.08 (1.6%) 82.85 (1.7%)
2.2% ( -1% - 5%) 0.000
BrowseDayOfYearSSDVFacets 4.01 (0.7%) 4.12 (0.4%)
2.7% ( 1% - 3%) 0.000
TermDTSort 60.83 (9.0%) 63.12 (11.6%)
3.8% ( -15% - 26%) 0.250
```
It does not show any slowdowns.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
