Re: No atime
Linus Torvalds (torvalds@cs.helsinki.fi)
Thu, 6 Jun 1996 13:53:47 +0300 (EET DST)
> Date: Thu, 6 Jun 96 00:13 BST
> From: Jamie Lokier <jamie@rebellion.co.uk>
>
> Another possibility is to not mark inodes dirty just because the atime
> field changes. Or mark them dirty in such a way that they don't need to
> get written out until the filesystem is unmounted. This would stop
> those laptop disks from spinning and probably be Posix compliant too (as
> long as the system is running).
>
> Stephen and I talked about doing this very thing at the Berlin Linux
> conference. I think this is something you can expect to see in the 2.1
> development cycle.
Umm. I actually _implemented_ it in something like five minutes, but decided we
don't want to have that kind of change just before 2.0...
The problem with the approach is that when you don't write them out, you end up
with them in memory taking up space, and then the "get_free_inode()" thing
might become much less efficient. So you have to flush them even at other times
than just unmount, and I never actually even bothered to reboot with my patches
in place, because this is exactly the kind of subtle problem that I _don't_
want to introduce right now..
Linus