[Nottingham] atime again...

[Martin]

Folks,

Why atime and does /anyone/ use it anywhere?


This has come up in discussion again and for whether we need bother
supporting it at all...


"atime" is a filesystem mount option that is the default and enables
recording the time when a file or directory is accessed ("access time").

For filesystem design, that is a very good idea to be included for if
such a thing might be useful.


However:

There is a big penalty in that you have the absurd situation that by
merely reading from a filesystem, you generate lots of random write
operations to update the filesystem to write back timestamps for
anything looked at.

That is such a overhead to slow things down that most distros now set as
a default "reltime" that only updates an atime only if the new timestamp
is more than 24 hours more recent. That greatly reduces the writeback
overhead for commonly access areas of the filesystem.

(For "read-only" mounted filesystems, atime is rendered useless in any
case because you cannot write back the updated access times.)


But...

Does anything or anyone make any use of atime other than for doing
filesystem access pattern analysis/forensics?

So is atime more of a filesystem debug feature rather than of real-world
use?...


The /only/ use case I know is where Mutt monitors atime for mbox files.
However, that is an old obsolete way of doing things. As far as I know,
there are *no other use cases* ...


Should we have noatime as the new default for mounting filesystems?


Or would that cripple how anyone might work?

Cheers,
Martin


(If you still do not know what atime is, then come to a meetup and all
can be elucidated ;-) Also, you need not worry about it at all! :-) )


-- 
- ------------------ - ----------------------------------------
-    Martin Lomas    - OpenPGP (GPG/PGP) Public Key: 0xCEE1D3B7
- martin @ ml1 co uk - Import from   hkp://subkeys.pgp.net   or
- ------------------ - http:// ml1 .co .uk/martin_ml1_co_uk.gpg