>> > Q2: I can't use " ufsdump 0cuf aix_box:/dev/rmt0 / " in this environment
>> > like I did from Solaris to remote Solaris tape drive, why ?
Because the "remote tape" protocol was poorly designed. That is, when
you specify host:device as the file for ufsdump, you are asking it to
use rsh protocol to invoke /usr/sbin/rmt on the remote machine.
Then ufsdump and rmt transfer data using a simple protocol (man rmt).
Unfortunately, rmt protocol was designed in the early BSD days, when
essentially all Unix systems that implemented rsh were running BSD
Unix. The designers of rmt took a shortcut--when they need to perform
a non-data operation on a tape (e.g., rewind, skip file, write tapemark,
etc.), they send the raw tape ioctl number to the remote host. This worked
fine between two identical BSD systems. However, as systems diverged from
BSD, the meanings of various ioctl numbers also diverged.
The result is that one vendor's dump programs might or might not work
with another vendor's rmt. In particular, AIX and Solaris are not
We had a kludge that we used. We backed up Solaris hosts to tapes on AIX
boxes. To make this work, we had a hacked version of rmt that mapped the
numbers provided by Solaris's ufsdump to the correct AIX ioctl. The source
is available from <ftp://quest.cc.purdue.edu/pub/aeh/rmt.tar>.
Nowadays, we use Amanda (www.amanda.org). One result is that hosts without
tape drives spool their dumps to disks on hosts with tapes. We no longer
Dale Talcott, Purdue University Computing Center