David Zuccaro <david.zuccaro@optusnet.com.au>
writes:

> On 21/08/15 12:24, Trent W. Buck wrote:
>>
>> By default mke2fs reserved 5% of the filesystem for uid 0.
>> This is also used as a staging area to reduce fragmentation.
>>
>> df does not include this area in the "free" size,
>> so if you fill a disk *as root*,
>> you must remove >5% for df to show any change.
>>
>> 30G is about 2% of 1.4T, so that's not enough.
>>
>> You can remove this reserved area using tune2fs,
>> but that is a bad idea.
>>
>>
>>
> Thanks Trent, that would explain the anomaly. I did not know that that
> the 5% reserved area was accessible to root (I was running the back up
> as root). Would it be ok to reduce the reserved area to say 1%?

It won't hurt.

Without reserved space, you will get more fragmentation (if you
repeatedly fill the filesystem).

BUT you're running the backup as root,
so it effectively ignores that reserved space ANYWAY.

e2fsck -f reports fragmentation as a percentage,
but I'm not sure how believable that number is.

                                 * * *

You could just put up with it.
IME fragmentation isn't a big deal on ext.

You could change the reserved user ID from 0 (using tune2fs),
so the root-owned rsync can't use the reserved space.

You could run a defrag program occasionally.
This mostly consists of writing a new copy of each file,
along these lines (untested, pre-coffee):

    nice ionice -c3 find -O3 /srv/backup -xdev -type f -links 1 -size +1M \
      -exec cp -v {} {}~ \; \
      -exec mv -v {}~ {} \;

...but it is easy to break unusual files (e.g. hard-linked, sparse)
without realizing until it's too late.

Last time I looked, all of the ext defrag programs were written by
recent Windows refugees and I didn't trust them to get edge cases right
any better than the above.

_______________________________________________
luv-main mailing list
luv-main@luv.asn.au
http://lists.luv.asn.au/listinfo/luv-main