>Date: Fri, 7 Nov 1997 12:56:49 -0800 (PST)
>From: Ross Harvey <ross@teraflop.com>

>> The block device interface to tapes hasn't really ever worked (the semantics
>> of implementing it are just too hairy, for one), and probably ought to just
>> be removed.
>> 
>> Jason R. Thorpe                                       thorpej@nas.nasa.gov

>It's doubtful for writes, but the block reads worked once, at least.
>One night around the end of the last ice age, just for fun I dd'ed a
>file system off a pdp-11 unix onto a 125 ips nine track. I mounted the
>block device read-only, took a deep breath, and typed "ls".

Yup. booting a readonly root from tape was *the* (or at least `a')
standard installation method for the early Pyramids (remember them?)
This was back in the mid-80s before CD-roms were widespread, let alone
affordable, and machines didn't have enough RAM for ramdisks.

There are a couple of ports from that period where this might still be
a useful option for installs where there's no network.  But Zip drives
are getting so cheap it's not a very compelling case -- except maybe
for machines that don't have SCSI.  

If the choice is root-on-tape versus writing a standalone copy or a
standalone restore, I'd go with the tape.

Besides, didn't write semantics work `properly' with some tapes, like
those hp300 block-addressable, expensive-medium, non-QIC quarter-inch tapes?