adriangb commented on PR #10020: URL: https://github.com/apache/arrow-rs/pull/10020#issuecomment-4622513297
@alamb I'm pretty convinced the benchmark regressions are just noise. From a logical standpoint this PR adds some indirection but does not really change the default code path. This should not have any overhead. Looking at some of the regressions in https://github.com/apache/arrow-rs/pull/10020#issuecomment-4617415597 `write Binary(100) delta_byte_array` is 60% slower but in https://github.com/apache/arrow-rs/pull/10020#issuecomment-4617641533 it is 7% slower. I think this noise needs a solution similar to https://github.com/apache/arrow-rs/pull/10068. These benches are all allocation heavy so any "bad luck" in ordering / timing of allocations (re-use vs. page fault) can result in big swings in results. My suggestion is that we don't over-index on this and move forward with this change if everything else looks good. -- 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]
