eldenmoon commented on PR #67551: URL: https://github.com/apache/doris/pull/67551#issuecomment-5709069223
Ran the full `^BM_Variant` suite (all 35 `BM_VariantCumulativeCompaction` + 20 `BM_VariantSparseImport` cases, default 1,000,000 rows) three times back to back on a shared 192-thread build host to record a baseline and check run-to-run noise. All 165 case-runs completed with `error_occurred: false`. Also moved the two benchmark files into `be/benchmark/variant/` (mirroring the existing `be/benchmark/parquet/` layout) and added `README.md` + `BASELINE.md` (build/run instructions and the full recorded numbers, including per-round data and the load-noise caveat below) — no benchmark logic changed, only `benchmark_main.cpp`'s include path. Headline (median of rounds 1+3, which ran under comparatively low background load — round 2 overlapped a load-average spike to ~349 on this box and is ~20-100% slower across the board; see `BASELINE.md` for the full breakdown): **`BM_VariantCumulativeCompaction`, `real_time` ms (lower is faster):** | Scenario | Median (n=10) | |---|---:| | Sparse16/FlatPhysicalColumns | 7,305 | | Sparse16/MixedPhysicalPlacement | 7,216 | | Sparse16/WholeVariant | 15,977 | | Full/WholeVariant | 16,154 | | Doc16/DocBuckets | 18,958 | | Doc16/DocBucketsMaterialized | 22,930 | | Doc16/WholeVariant | 17,200 | **`BM_VariantSparseImport`, `cpu_s_per_1m_rows` (lower is faster):** | Scenario | Median (n=10) | |---|---:| | MixedTypes/Writers1 | 10.855 | | NoArrays/Writers1 | 9.094 | | MixedTypes/Writers8 | 14.378 | | NoArrays/Writers8 | 10.671 | `Writers8` costing more than `Writers1`, and `NoArrays` coming in under `MixedTypes` in both writer modes, matches this PR's own description of the array-inference/Writers8 contention this benchmark targets. These absolute numbers are specific to this shared host and moment of contention — see `BASELINE.md`'s "Machine-load impact" section before using them as a regression/improvement reference. -- 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]
