I'm sure that this didn't used to happen (well it doesn't on SL3)...

$ cd /tmp
$ touch new-file
$ ls -al new-file 
-rw-r--r--  1 jp107 other 0 Oct  6 01:08 new-file
$ du -sk new-file 
4       new-file
$ echo > new-file
$ ls -al new-file 
-rw-r--r--  1 jp107 other 1 Oct  6 01:08 new-file
$ du -sk new-file 
8       new-file

if I make slightly bigger files then the 'extra' blocks allocated also
goes up (to 32K extra that I've seen).

$ uname -a
Linux hilus 2.6.9-42.0.2.EL #1 Tue Aug 22 16:56:02 CDT 2006 i686 i686 i386 GNU/Linux

but I also get the same behaviour with:

$ uname -a
Linux cingulum 2.6.9-42.0.2.ELsmp #1 SMP Tue Aug 22 18:36:24 CDT 2006 x86_64 x86_64 x86_64 GNU/Linux

On hilus I've got ext3 on LVM2 on a plain (IDE) disk, while on
cingulum it is ext3 on LVM2 on an external SCSI raid device (not that
that should matter).  I get the same behaviour with 

I don't really mind the files getting extra blocks 'allocated'
e.g. like the block-reservation patches I remember seeing floating
past several years ago, but this seems to actually count towards the
blocks used (and also user quotas etc).  Some user directories (and so
quota usage) went up by over 30% after copying to an SL4 box.

No unusual mount options are being used.  After a google search I
tried adding noreservation but that made no difference.

Am I just going mad or is there some option I'm completely
overlooking?  Or does everyone else just ignore it (disk is cheap...)

-- 
Jon Peatfield,  Computer Officer,  DAMTP,  University of Cambridge
Mail:  [log in to unmask]     Web:  http://www.damtp.cam.ac.uk/