jpountz commented on PR #12664:
URL: https://github.com/apache/lucene/pull/12664#issuecomment-1759536405

   `luceneutil` on `wikibigall` gave good results, better than I expected:
   
   ```
                               TaskQPS baseline      StdDevQPS 
my_modified_version      StdDev                Pct diff p-value
              HighTermDayOfYearSort      270.41      (2.0%)      255.63      
(2.3%)   -5.5% (  -9% -   -1%) 0.000
                    CountOrHighHigh       57.44     (16.1%)       54.37     
(11.0%)   -5.4% ( -27% -   25%) 0.220
                     CountOrHighMed       89.17     (15.9%)       84.60     
(11.1%)   -5.1% ( -27% -   25%) 0.237
                    CountAndHighMed      122.57      (3.4%)      120.69      
(4.4%)   -1.5% (  -8% -    6%) 0.212
                            Prefix3      105.15      (5.9%)      104.66      
(5.4%)   -0.5% ( -11% -   11%) 0.795
                           Wildcard      112.31      (3.2%)      111.94      
(3.4%)   -0.3% (  -6% -    6%) 0.758
                        CountPhrase        4.41      (2.2%)        4.40      
(4.2%)   -0.3% (  -6% -    6%) 0.776
                             IntNRQ      164.43     (13.4%)      163.96     
(12.7%)   -0.3% ( -23% -   29%) 0.945
                            Respell       73.48      (1.9%)       73.51      
(1.7%)    0.0% (  -3% -    3%) 0.944
                          CountTerm    17115.51      (4.7%)    17122.81      
(6.5%)    0.0% ( -10% -   11%) 0.981
                             Fuzzy2      126.72      (1.2%)      127.18      
(1.4%)    0.4% (  -2% -    2%) 0.380
                           PKLookup      224.32      (1.8%)      225.19      
(2.1%)    0.4% (  -3% -    4%) 0.532
                             Fuzzy1      148.99      (1.3%)      149.61      
(1.5%)    0.4% (  -2% -    3%) 0.345
                          LowPhrase       22.22      (3.9%)       22.38      
(3.6%)    0.7% (  -6% -    8%) 0.529
                            LowTerm     1050.65      (6.0%)     1058.66      
(4.2%)    0.8% (  -8% -   11%) 0.641
                          MedPhrase       69.20      (4.0%)       69.91      
(3.3%)    1.0% (  -6% -    8%) 0.377
                            MedTerm      611.39      (7.0%)      618.72      
(4.8%)    1.2% (  -9% -   13%) 0.525
                  HighTermMonthSort     5173.19      (2.6%)     5241.52      
(2.2%)    1.3% (  -3% -    6%) 0.084
                           HighTerm      425.31      (7.9%)      431.22      
(6.0%)    1.4% ( -11% -   16%) 0.532
                         HighPhrase       45.70      (5.1%)       46.39      
(4.3%)    1.5% (  -7% -   11%) 0.311
                   CountAndHighHigh       40.59      (4.0%)       41.33      
(5.1%)    1.8% (  -6% -   11%) 0.208
                          OrHighLow      554.87      (3.7%)      579.95      
(3.8%)    4.5% (  -2% -   12%) 0.000
                         OrHighHigh       47.17      (4.9%)       49.35      
(5.1%)    4.6% (  -5% -   15%) 0.003
                        AndHighHigh       69.41      (4.1%)       73.39      
(3.9%)    5.7% (  -2% -   14%) 0.000
                          OrHighMed      244.77      (3.4%)      261.52      
(3.3%)    6.8% (   0% -   13%) 0.000
                         AndHighMed      124.85      (3.4%)      134.47      
(3.4%)    7.7% (   0% -   15%) 0.000
                         AndHighLow     1059.79      (2.7%)     1152.15      
(2.2%)    8.7% (   3% -   14%) 0.000
   ```
   
   All `OrXY` and `AndYY` tasks show a good speedup with a p-value equal to 0. 
I'm pretty sure that the queries that show a regression are noise.


-- 
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

Reply via email to