[comp.unix.ultrix] DECNet-Ultrix 4.0 installation

osyjm@cs.montana.edu (Jaye Mathisen) (11/30/90)

Scenario...

  DS 5000 with full set of Ultrix including DECNET...

  DS3100 with RZ23, that mounts the 5000 /usr...  What's the proper
sequence/where do things go to put the ncp command to start the
executor?  If you put it after the ifconfig that's necessary to get
TCP/IP running so you can mount /usr from the 5000, then things seem
to hang forever...  So I says, How about putting ncp in /etc and firing
it up that way, but it seems to need some other files.  What is the
correct way to do this besides buying an RZ55 and having a full local
copy of Ultrix?

--
 Jaye Mathisen,sysmgr 410 Roberts Hall,Dept. of Computer Science
 Montana State University,Bozeman MT 59717         PHONE: (406) 994-{4780,3931} 

mellon@fenris.pa.dec.com (Ted Lemon) (11/30/90)

Ouch.   What it needs is /usr/lib/dnet.   The following crock might work:

!!!!!!!!!!!!!!!WARNING WARNING WARNING WARNING WARNING!!!!!!!!!!!!!!!

The following is probably a disgusting kludge, is definitely not
supported by DEC, and will probably cause both your machines to emit
sparks, melt their disks into useless sludge, and cause portions of
your network cabling to evaporate in protest.

It's also possible that I've forgotten something - I try to avoid
hacking on DECnet whenever possible.  If you choose to try the
instructions below, you're solely responsible for whatever cleanup
needs doing when the whole thing blows up in your face.

!!!!!!!!!!!!!!!WARNING WARNING WARNING WARNING WARNING!!!!!!!!!!!!!!!

	(1) Bring both machines down to single user mode.
	(2) Make a /dnet on both root partitions (the RZ23 and the
		DS5000's root).   Copy all the files in /usr/lib/dnet
		to both directories.
	(3) On the 3100, with /usr not mounted, create the directory /usr/lib.
	(4) Symbolically link /dnet to /usr/lib/dnet.
	(5) On the 5000, remove /usr/lib/dnet, and symlink /dnet to
		/usr/lib/dnet.
	(6) Reboot the 3100 into single user mode; mount /usr.
	(7) Change the 3100's DECnet address, node name and identification.
		ncp define executor address new.address
		ncp define executor node newnod
		ncp define executor identification "Hacked-off 3100"
	(8) Reboot the 3100.   ncp should be run before ifconfig, as
		you have already divined.

The problem with setting things up as you've done is that the node
name and address of the machine is stored in a file in /usr/lib/dnet.
Since you were mounting /usr on both the 5000 and the 3100, once you
ran ncp, both machines had the same ethernet address.   I think you
can imagine the implications of this.

Anyway, good luck!

			       _MelloN_

mamros@energy.irc.cbm.dec.com (Shawn Mamros) (11/30/90)

In article <MELLON.90Nov29233750@fenris.pa.dec.com>, mellon@fenris.pa.dec.com
(Ted Lemon) writes:
|> !!!!!!!!!!!!!!!WARNING WARNING WARNING WARNING WARNING!!!!!!!!!!!!!!!
|> 
|> The following is probably a disgusting kludge, is definitely not
|> supported by DEC, and will probably cause both your machines to emit
|> sparks, melt their disks into useless sludge, and cause portions of
|> your network cabling to evaporate in protest.
|> 
|> It's also possible that I've forgotten something - I try to avoid
|> hacking on DECnet whenever possible.  If you choose to try the
|> instructions below, you're solely responsible for whatever cleanup
|> needs doing when the whole thing blows up in your face.
|> 
|> !!!!!!!!!!!!!!!WARNING WARNING WARNING WARNING WARNING!!!!!!!!!!!!!!!

That goes double for this article here.

Now to the original posting...
In article <3016@dali>, osyjm@cs.montana.edu (Jaye Mathisen) writes:
|>> Scenario...
|>>   DS 5000 with full set of Ultrix including DECNET...
|>>   DS3100 with RZ23, that mounts the 5000 /usr...

First of all, just so you know, that configuration (root on local disk,
/usr over the network) isn't officially supported either.  I'm not trying
to criticize your use of it, nor am I saying you shouldn't use it (mainly
because I use it here ;-)... I'm just telling you what others have told me.

Now, if you are going to be running things that way, one thing that you
need to make sure of is that each machine has a separate /var area.
If /var is a soft link to /usr/var, then you're going to have lots of
problems with DECnet and other things too.  It's best to create a separate
/var partition on each machine and have /usr/var be a soft link to /var.

As Ted pointed out, the DECnet databases for each machine (including its
identity) live in /usr/lib/dnet.  But that's actually a soft link to
/usr/var/dnet.  So, if you have separate /var partitions, you're halfway
home...

Now, as far as starting DECnet goes, I have gotten it to work in that
configuration by running ifconfig to start IP, mounting /usr and then
using ncp to start DECnet.  I'm not sure why this works in the face of
the changing Ethernet address, but I've seen it happen (at least on
ULTRIX V3.1 and V4.0 on RISC machines).  No, it's not supported, and if
it breaks in a future release you're stuck, but that goes for everything
else in here too.

-Shawn Mamros
E-mail to: mamros@crl.dec.com
Disclaimer: No, I'm not an ULTRIX developer, just another "satisfied customer"

mellon@fenris.pa.dec.com (Ted Lemon) (12/01/90)

>Now, as far as starting DECnet goes, I have gotten it to work in that
>configuration by running ifconfig to start IP, mounting /usr and then
>using ncp to start DECnet.  I'm not sure why this works in the face of
>the changing Ethernet address [...]

The reason that Jaye's system froze completely was that both the
DS5000 and the DS3100 had the same ethernet address, meaning that
neither one could tell the difference between the other, and meaning
that both would respond to requests from any machine on the net
directed at the DS5000.

In the 3100's case, it would probably respond to its own nfs requests
with a message on the order of ``What in the blue blazes are you
talking about???''.  Major lossage.  It's also possible that the IP
code would hang in some less obvious way - I don't know enough about
the guts of it to know for sure.

Once you've fixed the problem of both machines having the same
ethernet address, you're left with the problem that the data in the
DS5000's arp cache is incorrect.  This will fix itself fairly quickly,
and the symptom will just be a long delay on the boot procedure on the
3100.

				_MelloN_