gf2121 commented on PR #12661: URL: https://github.com/apache/lucene/pull/12661#issuecomment-1760721571
I made an experiment that run luceneutil without `-r`, which means we read old format with [candidate code](https://github.com/apache/lucene/blob/92c13b8e59c5e4cf6de5e49863e0b102d5063987/lucene/core/src/java/org/apache/lucene/codecs/lucene90/blocktree/FieldReader.java#L110). The result looks good, proving that the regression indeed caused by `readMSBVLong`. ``` TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff p-value Prefix3 93.67 (3.8%) 91.90 (3.1%) -1.9% ( -8% - 5%) 0.085 OrHighMed 24.75 (4.7%) 24.44 (4.5%) -1.3% ( -10% - 8%) 0.391 AndHighLow 299.77 (4.3%) 297.03 (3.7%) -0.9% ( -8% - 7%) 0.474 LowTerm 274.06 (8.4%) 271.71 (4.4%) -0.9% ( -12% - 13%) 0.687 OrHighHigh 13.74 (3.6%) 13.63 (5.3%) -0.8% ( -9% - 8%) 0.561 HighPhrase 58.48 (2.8%) 58.05 (5.0%) -0.7% ( -8% - 7%) 0.568 OrNotHighMed 128.42 (4.4%) 127.52 (4.5%) -0.7% ( -9% - 8%) 0.622 Fuzzy1 50.71 (2.3%) 50.37 (2.8%) -0.7% ( -5% - 4%) 0.402 MedPhrase 68.38 (5.1%) 68.08 (4.4%) -0.4% ( -9% - 9%) 0.769 AndHighHigh 10.35 (2.8%) 10.33 (2.8%) -0.2% ( -5% - 5%) 0.844 Fuzzy2 28.57 (2.8%) 28.53 (2.0%) -0.2% ( -4% - 4%) 0.832 LowPhrase 8.81 (3.7%) 8.80 (2.6%) -0.1% ( -6% - 6%) 0.887 OrNotHighHigh 111.28 (5.4%) 111.31 (5.0%) 0.0% ( -9% - 10%) 0.983 Respell 25.65 (2.5%) 25.68 (2.3%) 0.1% ( -4% - 5%) 0.893 OrHighNotLow 147.62 (5.2%) 147.81 (5.6%) 0.1% ( -10% - 11%) 0.939 HighTerm 229.21 (4.5%) 229.52 (4.9%) 0.1% ( -8% - 9%) 0.927 MedSloppyPhrase 8.03 (2.3%) 8.04 (3.1%) 0.2% ( -5% - 5%) 0.797 PKLookup 97.60 (3.7%) 97.84 (4.1%) 0.2% ( -7% - 8%) 0.845 OrHighNotHigh 114.38 (4.4%) 114.78 (5.1%) 0.3% ( -8% - 10%) 0.818 LowIntervalsOrdered 3.93 (2.2%) 3.95 (2.5%) 0.4% ( -4% - 5%) 0.628 OrNotHighLow 330.79 (4.4%) 332.00 (3.9%) 0.4% ( -7% - 8%) 0.778 LowSloppyPhrase 6.52 (2.9%) 6.55 (4.6%) 0.5% ( -6% - 8%) 0.690 Wildcard 89.99 (3.5%) 90.55 (2.9%) 0.6% ( -5% - 7%) 0.545 LowSpanNear 22.40 (2.3%) 22.54 (1.7%) 0.6% ( -3% - 4%) 0.335 HighSpanNear 0.78 (3.0%) 0.79 (2.4%) 0.7% ( -4% - 6%) 0.405 MedSpanNear 8.11 (3.2%) 8.19 (3.0%) 0.9% ( -5% - 7%) 0.345 HighSloppyPhrase 5.66 (3.6%) 5.71 (4.6%) 1.0% ( -6% - 9%) 0.437 AndHighMed 51.51 (4.1%) 52.04 (3.9%) 1.0% ( -6% - 9%) 0.410 OrHighLow 130.83 (3.1%) 132.29 (2.8%) 1.1% ( -4% - 7%) 0.233 OrHighNotMed 142.66 (5.4%) 144.34 (5.4%) 1.2% ( -9% - 12%) 0.488 MedTerm 340.68 (3.7%) 344.79 (4.1%) 1.2% ( -6% - 9%) 0.329 HighIntervalsOrdered 6.20 (6.4%) 6.28 (5.6%) 1.2% ( -10% - 14%) 0.520 MedIntervalsOrdered 6.07 (9.6%) 6.18 (8.4%) 1.9% ( -14% - 21%) 0.505 ``` -- 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
