Rich-T-kid commented on PR #26977: URL: https://github.com/apache/datafusion/pull/26977#issuecomment-6732226494
Re-ran the focused ClickBench comparison on the rebased branch using released Arrow 60.1.0. Setup: - base: `b8534c7fd836f23e99f3f95fccea53cd7e167bff` - candidate: `75e94abfe45e7266ae1cd234154127189163c031` - partitioned ClickBench dataset, 12 partitions, 32 GiB memory limit - `datafusion.execution.parquet.enable_rle_to_dictionary=true` on both sides - four balanced base/candidate/candidate/base runs, 50 iterations per query per run - separate Cargo target directories; base binary SHA-256 `e906be784c24dff91285eb4a03b089b03769756d426c5a9e03af347158c97ac9`, candidate `70c7aea70cffdc965626ded3aaeafc65ff7065bc305b780a4cca80658e8f2343` Pooled medians: | Query | Main | PR | Change | | --- | ---: | ---: | ---: | | Q10 | 176.26 ms | 59.41 ms | -66.3% / 2.97x faster | | Q11 | 188.13 ms | 70.61 ms | -62.5% / 2.66x faster | The profiles now enter the Parquet dictionary decoder directly instead of spending roughly half of CPU time in `DecoderProjection` / `pack_byte_to_dictionary` rebuilding keys row by row. -- 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]
