The noatime Mount Option – Considered Harmful

Apparently Linus hates noatime as much as I do:

In a recent lkml thread, Linus Torvalds was involved in a discussion about mounting filesystems with the noatime option for better performance, “‘noatime,data=writeback’ will quite likely be *quite* noticeable (with different effects for different loads), but almost nobody actually runs that way.” He noted that he set O_NOATIME when writing git, “and it was an absolutely huge time-saver for the case of not having ‘noatime’ in the mount options.

Mounting your drives with noatime can be a huge performance boost. In past benchmarks I’ve seen it save as much as 8% but of course your milage may vary.

  1. NOTE: I should have rewritten this as ‘atime’ consider harmful. Ug.

  2. sykosoft

    I’ve seen even greater performance. I had a web server when I started as a Sr Systems Engineer with a company that had constant load averages of 20-30. Not only that, it was seeing 20-30mbit of traffic, all on small htm, jpg, and php files. Remounting it with noatime decreased the load averages to 0.50, and allowed me to increase the traffic the server was capable of by 500%.

    Nowadays, we simply use reiserfs with noatime.


  3. nobody

    Your description was very confusing and messed up. If I didn’t read through the link I would think Linus didn’t like noatime like you said.

    He doesn’t like atime!

%d bloggers like this: