Rich-T-kid commented on PR #10220: URL: https://github.com/apache/arrow-rs/pull/10220#issuecomment-4818934211
🤔 i've been thinking about changing the benchmarks to avoid having tokio polling be a factor. It may be more useful to benchmark the top level entry functions I.E - encode path : `encode_batch()` - decode path: `extract_message()` directly This would avoid the tokio run time playing such a large role in polling the streams, which I think is causing large variances between benchmarks. For example ``` decode/fixed/65536x8 7.99 4.5±0.49ms 3.5 GB/sec 1.00 563.9±12.87µs 27.7 GB/sec decode/nested/65536x8 2.38 15.3±2.01ms 2.6 GB/sec 1.00 6.4±1.36ms 6.1 GB/sec ``` the decode path wasnt touched in this PR but its showing a drastic change we have no control over the tokio run time so it makes sense to avoid it in the benchmarks if we can. @alamb do you have any opinions on this? -- 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]
