I rescue lots of kit from the tip. Presently, I am using 8 spindle
servers to replace (reimplement) my SAN appliances (the appliances
are closed source; file servers can be OPEN source!).

I was in the process of duplicating a spindle:
# cd /0; tar cpf - * | (cd /4; tar xpf - )
and belatedly decided to make some changes to the structure of the
source filesystem. So, killed off that job and "rm -r /4/*".

This always mucks with the free block count which is annoying;
it's nice to df(1) and see identical block counts after a copy so:
# umount /4; newfs -U -j /dev/da4p1; mount /4
makes the disk look pristine.

OK, restart the copy:
# cd /0; tar cpf - * | (cd /4; tar xpf - )
And, lets see what sort of throughput we're getting:
# iostat -c5 da0
WTF? *nothing*??? Literally *0* MB/s!
# iostat -c5 da4
Hmmm... 150MB/s. That explains why the disk activity indicator for
drive 4 is blinking like crazy! And, disk 0 is solid.

Open another session and I can see space being consumed on drive 4.
And, the actual contents from drive 0 appearing as time passes!

After some 3 minutes, the activity indicator for disk 0 starts
blinking...

Ah! Cached files from the prior "copy". Some 28GB of cached
files! Well, it's good to see that the machine is making use
of that memory (96GB). I wonder how much is set aside for the disk
cache vs. network and other kernel buffers, etc.

Quite a disturbing experience (did I type the command incorrectly?)

--- PyGate Linux v1.5.10
* Origin: Dragon's Lair, PyGate NNTP<>Fido Gate (3:633/10)