Re: Atime behavior. Was: Re: 3.0 wishlist

linux kernel account (linker@nightshade.z.ml.org)
Mon, 19 Jan 1998 20:22:34 -0500 (EST)


I had the same idea of lowering atime res... I though no one would agree
on it so it was pointless, I never tought of a mount (or format) option.

On Mon, 19 Jan 1998, John Kodis wrote:

> On Mon, Jan 19, 1998 at 10:00:05AM +0100, Mikael Abrahamsson wrote:
>
> > I would like to be able to set noatime as a mount option so I can
> > choose what filesystems I want noatime on and not.
>
> Rather than just setting atime to on or off, it would be useful to
> allow an atime delta, to specify how accurate an atime value was
> required. For example, if a file system were mounted with
> adelta=86400, the atime value of the files on this file system would
> only be updated if the current atime value is more than 86400 seconds
> (i.e., one day) ago.
>
> This would avoid the need to perform an inode write each time a file
> is accessed, while still providing a good indication of which files
> can be safely rolled off to a tape archive after having not been
> accessed for 6 months. Setting adelta=1 would yield the current
> behavior, and adelta=0 could be interpreted as 'no atime'.
>
> VAX/VMS provided such a capability, so there's some past practice with
> such a scheme. On VMS, IIRC, the adelta value was settable on a
> file-by-file basis, with newly created files inheriting a default from
> their parent directory. Given the current inode space shortage, this
> would have to wait for ext3. However, providing this feature on a
> file system basis seems like something that could be added to the
> existing no-atime code fairly easily.
>
> -- John Kodis.
>