kosiew commented on PR #15648: URL: https://github.com/apache/datafusion/pull/15648#issuecomment-2796382863
Findings after trying many optimisation attempts: 1. benchmark results are not stable. At the µs scale, many factors(besides the code itself) can swing the results. 2. only some changes are statistically significant (p < 0.05) and I [modified my script to retrieve only significant results](https://github.com/apache/datafusion/issues/15636#issuecomment-2795739270) Below are some of the significant (unstable) changes, comparing this branch against `main`: <img width="468" alt="image" src="https://github.com/user-attachments/assets/f29fce9f-481b-4061-b40f-c74fff9fe07b"; /> <img width="468" alt="image" src="https://github.com/user-attachments/assets/7bc415ff-0df9-45b5-ba70-643643b87ffe"; /> <img width="468" alt="image" src="https://github.com/user-attachments/assets/8deb0644-8884-4b0b-81a2-b603fe9be250"; /> As an illustration of the unstable benchmark results, I ran cargo bench on `main` to compare against a saved `main` baseline and obtained these significant changes: <img width="510" alt="image" src="https://github.com/user-attachments/assets/914700d3-9cbb-4b7e-be16-2a8caac20d14"; /> <img width="495" alt="image" src="https://github.com/user-attachments/assets/5bd08dd2-4fdd-4c6c-9470-19b4bb1783e2"; /> -- 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: github-unsubscr...@datafusion.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: github-unsubscr...@datafusion.apache.org For additional commands, e-mail: github-h...@datafusion.apache.org
