niyue commented on pull request #11588: URL: https://github.com/apache/arrow/pull/11588#issuecomment-959249470
> Probably not. AFAIU, @niyue's complaint is more about the additional I/O (which would happen in the background, but may still hamper performance of other concurrent I/Os).. Correct. In my test, it is more about unnecessary IO. You can tell from @cyb70289's test as well, 800MB unnecessary IO was performed, if the disk has 200MB/s read bandwidth, it will fully occupy the disk IO for 4 seconds, which make other disk IO not possible. Although I don't perform any serious memory related testing for this madvice, I feel the program using mmap + random access pattern + default IO advice could affect other applications' performance as well. The program essentially reads lots of never used data into page cache in this case, and OS may have to replace existing data in page cache with these never read data, which will cause other applications to perform worse in the end. -- 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]
