kkewwei commented on issue #16626: URL: https://github.com/apache/lucene/issues/16626#issuecomment-5535560897
I add msm with sort tasks: https://github.com/mikemccand/luceneutil/commit/03d3ac2bc8c53e0c3d5770fb07ca6f44d31d1132 After 19 benchmark rounds, the final results are: ``` TaskQPS baseline StdDevQPS my_modified_version StdDev Pct diff p-value MSM16M15HighLowDaySort 239.70 (9.3%) 181.68 (4.5%) -24.2% ( -34% - -11%) 0.000 MSM5M3MediumDaySort 15.25 (4.9%) 14.97 (4.1%) -1.9% ( -10% - 7%) 0.194 MSM8M2MediumDaySort 7.59 (3.9%) 7.47 (5.3%) -1.5% ( -10% - 7%) 0.313 MSM16M2MediumDaySort 1.85 (3.1%) 1.83 (3.4%) -0.7% ( -6% - 5%) 0.496 MSM16M15LowDaySort 233.17 (7.7%) 232.04 (4.4%) -0.5% ( -11% - 12%) 0.806 MSM5M2HighDaySort 1.70 (3.2%) 1.69 (2.2%) -0.1% ( -5% - 5%) 0.890 PKLookup 168.96 (8.4%) 168.76 (9.0%) -0.1% ( -16% - 18%) 0.966 MSM5M3HighDaySort 2.48 (4.3%) 2.50 (4.5%) 0.6% ( -7% - 9%) 0.679 MSM16M2HighDaySort 0.71 (2.9%) 0.71 (2.3%) 0.7% ( -4% - 6%) 0.398 MSM8M2HighDaySort 1.07 (3.4%) 1.09 (2.7%) 1.2% ( -4% - 7%) 0.233 MSM5M3HighLowDaySort 24.12 (5.9%) 24.49 (4.1%) 1.6% ( -8% - 12%) 0.334 MSM8M7HighLowDaySort 167.15 (4.4%) 174.85 (4.3%) 4.6% ( -3% - 13%) 0.001 MSM8M7LowDaySort 339.61 (4.7%) 367.36 (4.5%) 8.2% ( -1% - 18%) 0.000 MSM5M2MediumDaySort 39.03 (5.6%) 42.39 (4.8%) 8.6% ( -1% - 20%) 0.000 MSM5M4HighDaySort 13.34 (5.7%) 14.72 (2.9%) 10.3% ( 1% - 20%) 0.000 MSM5M4LowDaySort 430.93 (6.4%) 496.01 (8.0%) 15.1% ( 0% - 31%) 0.000 MSM5M3LowDaySort 358.04 (6.9%) 419.98 (4.8%) 17.3% ( 5% - 31%) 0.000 MSM3M2LowDaySort 509.79 (6.0%) 601.72 (6.5%) 18.0% ( 5% - 32%) 0.000 MSM3M2MediumDaySort 62.26 (4.8%) 76.27 (4.2%) 22.5% ( 12% - 33%) 0.000 MSM3M2HighDaySort 52.29 (4.7%) 66.02 (4.9%) 26.2% ( 15% - 37%) 0.000 MSM3M2HighLowDaySort 44.56 (5.9%) 65.51 (7.6%) 47.0% ( 31% - 64%) 0.000 MSM5M4MediumDaySort 8.91 (6.1%) 13.78 (3.2%) 54.7% ( 42% - 68%) 0.000 MSM8M7MediumDaySort 7.36 (7.5%) 11.88 (4.7%) 61.5% ( 45% - 79%) 0.000 MSM8M7HighDaySort 1.51 (5.9%) 2.45 (5.6%) 62.6% ( 48% - 78%) 0.000 MSM16M15MediumDaySort 16.27 (9.1%) 26.47 (6.6%) 62.6% ( 42% - 86%) 0.000 MSM5M4HighLowDaySort 9.38 (6.2%) 15.28 (5.3%) 62.9% ( 48% - 79%) 0.000 MSM16M15HighDaySort 2.95 (11.1%) 5.03 (6.3%) 70.6% ( 47% - 99%) 0.000 ``` Most cases show significant improvements, suggesting the optimization is worthwhile. The only exception is `MSM16M15HighLowDaySort`. Next, I plan to: 1. Investigate how to optimize `WANDScorer` to support this feature. 2. Identify the cause of the performance regression in `MSM16M15HighLowDaySort`. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
