gf2121 commented on pull request #2113: URL: https://github.com/apache/lucene-solr/pull/2113#issuecomment-745134305
> @dweiss Thank you very much for the guiding of how to do a benchmark on Lucene, the lueneutil tool is really nice! > I repeatedly execute the wikimedium1m 20 times, the following is the result of the last iter. I guess it indicates that the reading performance is stable overall :) > > ``` > TaskQPS baseline StdDevQPS my_mod_ver StdDev Pct diff p-value > HighTermMonthSort 405.40 (11.6%) 390.20 (11.1%) -3.7% ( -23% - 21%) 0.295 > PKLookup 131.35 (4.5%) 128.26 (6.5%) -2.4% ( -12% - 9%) 0.183 > Respell 88.96 (8.0%) 86.91 (8.7%) -2.3% ( -17% - 15%) 0.383 > AndHighLow 611.88 (5.6%) 603.60 (9.2%) -1.4% ( -15% - 14%) 0.575 > BrowseDayOfYearTaxoFacets 21.70 (7.7%) 21.43 (9.1%) -1.3% ( -16% - 16%) 0.630 > HighSloppyPhrase 74.97 (4.8%) 74.10 (8.6%) -1.2% ( -13% - 12%) 0.601 > BrowseMonthSSDVFacets 91.65 (4.8%) 90.66 (5.8%) -1.1% ( -11% - 9%) 0.519 > MedPhrase 110.37 (8.8%) 109.19 (7.3%) -1.1% ( -15% - 16%) 0.674 > OrHighMed 167.04 (8.9%) 165.47 (7.6%) -0.9% ( -16% - 17%) 0.718 > OrHighHigh 86.33 (9.1%) 85.61 (7.9%) -0.8% ( -16% - 17%) 0.755 > LowSpanNear 156.83 (7.1%) 155.74 (6.5%) -0.7% ( -13% - 13%) 0.746 > Wildcard 139.21 (7.0%) 138.29 (9.8%) -0.7% ( -16% - 17%) 0.805 > BrowseDayOfYearSSDVFacets 78.95 (5.0%) 78.52 (5.8%) -0.5% ( -10% - 10%) 0.753 > Fuzzy1 76.09 (9.7%) 75.84 (7.7%) -0.3% ( -16% - 18%) 0.905 > MedSloppyPhrase 48.37 (5.9%) 48.22 (5.1%) -0.3% ( -10% - 11%) 0.859 > IntNRQ 310.73 (10.5%) 310.11 (12.2%) -0.2% ( -20% - 25%) 0.955 > HighTerm 759.07 (7.3%) 757.84 (12.1%) -0.2% ( -18% - 20%) 0.959 > BrowseMonthTaxoFacets 24.17 (9.5%) 24.19 (10.0%) 0.1% ( -17% - 21%) 0.984 > HighIntervalsOrdered 121.98 (5.8%) 122.10 (7.9%) 0.1% ( -12% - 14%) 0.964 > LowSloppyPhrase 188.90 (7.8%) 189.42 (5.7%) 0.3% ( -12% - 14%) 0.898 > AndHighMed 418.22 (9.0%) 420.49 (9.7%) 0.5% ( -16% - 21%) 0.855 > MedTerm 874.56 (7.6%) 880.02 (10.9%) 0.6% ( -16% - 20%) 0.833 > MedSpanNear 378.96 (7.2%) 381.70 (9.2%) 0.7% ( -14% - 18%) 0.781 > Fuzzy2 25.74 (9.8%) 25.97 (10.9%) 0.9% ( -17% - 23%) 0.777 > AndHighHigh 126.51 (6.1%) 127.75 (8.5%) 1.0% ( -12% - 16%) 0.676 > HighPhrase 165.71 (8.2%) 167.64 (11.3%) 1.2% ( -16% - 22%) 0.708 > BrowseDateTaxoFacets 21.40 (10.4%) 21.69 (10.0%) 1.3% ( -17% - 24%) 0.682 > LowTerm 1032.43 (8.0%) 1049.57 (10.1%) 1.7% ( -15% - 21%) 0.566 > OrHighLow 228.54 (5.0%) 232.36 (8.5%) 1.7% ( -11% - 15%) 0.446 > HighSpanNear 93.72 (7.4%) 95.46 (7.3%) 1.9% ( -12% - 17%) 0.427 > HighTermDayOfYearSort 380.08 (12.2%) 387.47 (9.4%) 1.9% ( -17% - 26%) 0.573 > LowPhrase 137.13 (8.2%) 140.30 (6.2%) 2.3% ( -11% - 18%) 0.314 > Prefix3 290.55 (11.0%) 299.98 (12.9%) 3.2% ( -18% - 30%) 0.390 > ``` Though this result looks fine, i'm still a bit worried about the cost of reading array in some cases, for example, in a different java version. So i updated this PR again to make sure that we only read arrays when the index is a variable. In other case, we still read final longs. There are only 14 lines more than the origin code, which i think can be acceptable :) ---------------------------------------------------------------- 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: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org