When the backup moves the files from one server to the other the results from df change every few seconds in an impossible manner. The source host is running rsync. On the destination host I'm running the following command every few seconds:

  echo `date` `df|grep md0`

Results are below:

Sat Jun 29 23:57:12 CEST 2013 /dev/md0 4326425568 579316100  3527339636 15% /MD0
Sat Jun 29 23:57:14 CEST 2013 /dev/md0 4326425568 852513700  3254142036 21% /MD0
Sat Jun 29 23:57:15 CEST 2013 /dev/md0 4326425568 969970340  3136685396 24% /MD0
Sat Jun 29 23:57:17 CEST 2013 /dev/md0 4326425568 1255222180 2851433556 31% /MD0
Sat Jun 29 23:57:20 CEST 2013 /dev/md0 4326425568 1276006720 2830649016 32% /MD0
Sat Jun 29 23:57:24 CEST 2013 /dev/md0 4326425568 1355440016 2751215720 34% /MD0
Sat Jun 29 23:57:26 CEST 2013 /dev/md0 4326425568 1425090960 2681564776 35% /MD0
Sat Jun 29 23:57:27 CEST 2013 /dev/md0 4326425568 1474601872 2632053864 36% /MD0
Sat Jun 29 23:57:28 CEST 2013 /dev/md0 4326425568 1493627384 2613028352 37% /MD0
Sat Jun 29 23:57:32 CEST 2013 /dev/md0 4326425568 615934400  3490721336 15% /MD0
Sat Jun 29 23:57:33 CEST 2013 /dev/md0 4326425568 636071360  3470584376 16% /MD0

As you can see I start from USE of 15% and after 15 seconds I'm at 37% (I don't need to mention that the backup can not copy this huge amount of data in such a short time). After ~20 seconds the cycle closes. I'm again roughly at the same usage as earlier. The value is reasonable, ca. 35 Mb were copied.

Can somebody explain to me what is going on? Does df only make an estimation of usage instead of used value?