> The 20011221-1.5ZA snapshot appears to be too corrupt to use. Newfs
> is known to be broken (both on the install floppies and in the main
> system), dump has blown up for me, and god knows what else could
> go wrong.

There is something I've not seen from everyone else .. isn't 1.5ZA
considered experimental??

I run NetBSD on three of my production DEC 5305s and reserved the
fourth to run 1.5ZA because I consider that kernel to be very much
experimental and I wouldn't run it in a production environment.

I did have the same problem installing it with the floppies and 
actually went in with 1.5.2 floppies to do the newfs and such .. 
there was a thread a few weeks ago pointing out this problem, but
apparently no one addressed it.  Again, being experimental code,
I wasn't surprised, but because I wanted to run an SMP kernel I
got around the broken newfs with the 1.5.2 floppies.

I won't tell you what you should and shouldn't do, but you should
just update from the 1-5 Release directory .. There you can find
the 1.5.3_alpha kernel and friends.

> Leaving this kind of thing around is problamatic. For example, a
> recent customer of mine and I lost about eight hours trying to get
> this snapshot installed on one of this machines, at a cost to him
> of about $1200. Needless to say, we're running Linux on those alphas
> now, and he will never touch NetBSD on any machine ever again.

If he knows anything about UNIX he'll probably come screaming back.

I gave linux way too many second chances .. from 0.99 to 2.4.x and
I will never touch it again.

I still kick myself for ditching my AT&T 3b2/522 machines and SystemVr3.2.3
to give linux a shot on PCs .. that was 1997.  I went from hundreds of
days of uptime to about 15 days with an occasional max of about 30 days.
Sure they were faster (though at the time, not by too much) but its the
same story with the LISP hackers that had their Symbolics LISP Machines
traded in for SUNs.  They boot fast, and they love to show you how fast
they boot, but when it came down to getting work done they didn't perform.

I've already posted the reasons why I wasn't using NetBSD at the
time but I guess it won't hurt again (though I was already running it
on another project site since at least 1995 along with a handful of
other machines I was administering to).  I was dealing with a userbase
of over 10,000 people .. BSD style hashed password file building was
just way too slow and linux used the SystemV shadow suite .. that was
the only reason why.

Stephen