zagto commented on PR #12957: URL: https://github.com/apache/arrow/pull/12957#issuecomment-1120009112
I now run ran one of the most-affected benchmarks - [tpch, arrow, parquet, memory_map=False, TPCH-15, scale_factor=1, R](https://conbench.ursa.dev/compare/benchmarks/3c980569f8be4d85a8d54e0ffb93e93c...ed4d1a2fb6a346478f3467cbe308c895/) - and saw no such difference on my machine. (provided R picked up the changed C++ library correctly, I'm not very familar with it) Going trough the benchmark runner results a bit more I noticed that all the extreme permanent changes are R TCPH benchmarks. For example [this](https://conbench.ursa.dev/compare/benchmarks/925f7fb18a244281880c996eefc42ba3...a2aa39dec1bf4bb284bbbbbcc04ecd2f/) and [this](https://conbench.ursa.dev/compare/benchmarks/e1ac2cb7e24a4d7d9193294a1d999fb2...ce9a5035803444b79fab77045c9bb786/) are the worst affected Python and JS benchmarks. Also some TCPH benchmarks see an extreme improvement. Overall this gives me the impression the performance change may be unrelated to this commit. I wonder if it is possible to run the baseline again on conbench to isolate this? [output-TPCH15.txt](https://github.com/apache/arrow/files/8643616/output-TPCH15.txt) -- 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]
