Changing DECnet address for nodes in a cluster

[Ken USA]

Hi - We have a cluster of 3 nodes - NODEA, NODEB and NODEC. All three are Alpha boxes - with "A" and "B" running OpenVMS 8.2 and "C" running 7.3-2. "A" and "B" are running DECnet IV and "C" has DECnet V.

The DECnet address for "A" and "B" are 50.268 and 50.269. The DECnet address for "C" is 63.1022.

We are able to connect from node A to B and from B to A by doing SET HOST. We are not able to connect to/from "C".

Also, there are several other Alpha machines in the building. We are not able to connect to/from these machines from/to any of the machines in our cluster. All alphas in the building are running DECnet V and have a decnet address of 4.xxx.

We are thinking of changing SCSSYSTEMID for all three machines in our cluster so that we have a DECnet address of 4.nnn for all of them.

What's the best way to do this without bringing the whole cluster down? If we have to bring one node down, that's okay but would like to avoid bringing the entire cluster (all 3 nodes) down.

Any help will be highly appreciated.

Thanks in advance.

-Ken

[Robert Gezelter · Accepted solution]

Ken,

You will have to change SCSSYSTEMID and the DECnet node numbers simultaneously. This is not a large problem. However, you will have a period when you will have incorrect translations from DECnet names to numbers.

It goes without saying that BEFORE doing this, I would clone the system roots and make the change on one of the copies. If something critical has a problem, it is thus safer to revert to the current configuration.

- Bob Gezelter, http://www.rlgsc.com

[Hoff]

The SCSNODE and SCSSYSTEMID are a pair, and all cluster member nodes will remember it for the lifetime of the cluster. They won't let a node join the cluster with a conflicting SCS address or a conflicting SCS name.

You can implement DECnet area routers for the connections (each DECnet area requires an area router for communications outside the area, and area routers connect to other area routers to allow packets from one network area to the next.)

To change SCSSYSTEMID, you have two choices: change both the SCSNODE and the SCSSYSTEMID and reconfigure DECnet to match the new address and new name (which looks like a new node), or shut down the whole cluster and reset the SCSSYSTEMID and the DECnet address.

If you need to keep some or all of the cluster nodes up, adding DECnet area routers (either adding the routers, or configuring existing DECnet hosts into area routers) will permit this. Otherwise, you're rebooting the cluster. Or you're adding the nodes back in under new names and new addresses.

If you wish to consider a cluster node name change, the list of affected components is included in the OpenVMS FAQ -- there's a section in the FAQ listing the various hunks of OpenVMS and of some LPs that can need to be changed if the node name changes. The FAQ home is at http://www.HoffmanLabs.com/vmsfaq/

Stephen Hoffman
HoffmanLabs

[Ken USA]

I need little help here with the commands. I am from the Applications team and we got two new nodes "A" and "B" to install our app and were asked to put the node "C" in the cluster.
Was told to change the SCSSystemId and DECnet address. Do I change SCSSystemId using SYSGEN or SYSMAN? Help with the commands to enter would be highly appreciated.

[Ken USA]

Hi - I responded before reading the last response.

>>> You can implement DECnet area routers
>>> for the connections (each DECnet area
>>> requires an area router for
>>> communications outside the area, and
>>> area routers connect to other area
>>> routers to allow packets from one
>>> network area to the next.)

This may be a better approach. We don't have to change SCSSystemId and DECnet address. We just need an ability to connect to other machines using "set host".

How could this be done?

Thanks,
-Ken

[Hoff]

>>>
need little help here with the commands. I am from the Applications team and we got two new nodes "A" and "B" to install our app and were asked to put the node "C" in the cluster.
Was told to change the SCSSystemId and DECnet address. Do I change SCSSystemId using SYSGEN or SYSMAN? Help with the commands to enter would be highly appreciated.
<<<


Make the changes in MODPARAMS.DAT, then AUTOGEN the box. See the OpenVMS FAQ for other stuff that can be associated with the node name.


>>> area routers connect to other area
>>> routers to allow packets from one
>>> network area to the next.)

>>>
This may be a better approach. We don't have to change SCSSystemId and DECnet address. We just need an ability to connect to other machines using "set host".

How could this be done?<<<

That depends on what hardware and software you have on-site.

I'll assume you have an area router for 4 and an area router for 50, so load up some rinky-dink VAX with a base OpenVMS license and a DECnet DVNETRTG routing license (or a Cisco or similar switch (and router) can likely provide this area routing, too) and set up an area router for area 63. This area router should be adjacent to the area routers for areas 4.* and 50.*.

IIRC, host-based DECnet phase IV routing was never officially supported on OpenVMS Alpha. DECnet-Plus Phase V host-based routing arrived at V7.1 or so, so converting your 63.* existing node that is a potential option.

[Hoff]

You can probably also use a VAX emulation with the base license and DVNETRTG, too. (Oops. Sorry Stanley. :-) ) Conceivably running on a Microsoft Windows box, maybe on Mac OS X box via SIMH, or on an Alpha box. Make sure you have enough bandwidth through the emulation, if you choose this route.

[Robert Gezelter]

Ken,

I agree with Hoff that a VAX simulator with DECnet Phase IV routing would solve the problem, I would also consider it a challenge for a beginning system manager.

As Hoff reminds, the cluster will "remember" the SCSNODE and SCSSYSTEMID, but you can likely accomplish things with little disruption.

However, I would advise that consideration be given as to whether someone with experience in cluster reconfiguration (such as Hoff or myself) be contacted. [disclosure: my consulting practice can provide this service; as can Hoff's].

WADU, learning cluster system management on a production cluster is a challenge, and can lead to unfortunate circumstanes (e.g., downtime). This is not a difficult transitio, but caution is advised.

- Bob Gezelter, http://www.rlgsc.com

[Jan van den Ende]

Ken,

As _I_ do not have any potential (financial) benefits, I feel I am a better unbiased source of advice.
But really, _DO_ consider the (paid) help of Hoff, or Bob, or ... (there are many more).
But do your employer (or whatever role the system owner holds) a BIG favor:
GET EXPERIENCED ASSISTANCE before messing with a production system!
It is WAY OUT the cheapest thing!

(and insist on watching him/her continously, and explain the WHY's; so next time YOU are the experienced one that can be for hire)

Bottom line: have any fun you like on a sandbox, but KNOW what you do on production systems!

hth

Proost.

Have one on me.

jpe

[Anton van Ruitenbeek]

Ken,

At a point you are mentioning to have routed DECNet. Not all routers nowadays do support DECNet. The best there is is DigitalNetworks and not C...co. But anyway, if you want to go out of the box using routers it's also a possibility to use DECNet over IP (free ofcourse). You still need to now how to configure DECNet (if you need this ofcourse) but you can now use a lot of routers who do not understand anything about DECNet.
If you are clustering you still need support of SCS ofcourse .

AvR

[Wim Van den Wyngaert]

Our routers only route decnet4. So, between nodes on the lan it works fine but when going over a router, only nsp addresses are used. Be sure to define addresses for decnet4 (nsp) if you want to be sure that all communications work well.

Wim

[Colin Butcher]

Hello,

A diagram would be most helpful. I'd guess that each machine has more than one NIC and that the Phase V machine has Phase IV addressing enabled on only one of them, not on all of them. However, you do need to be very careful to make sure that you don' end up with a duplicate MAC address.

The SCSSYSTEMID / DECnet Phase IV address should not be relevant if they're all on the same LAN / VLAN as the "on ethernet" bit will be set and you get direct communication without needing an area router.

If you have time for lighter reading than the Phase V doc set, try here: http://www.downloads.xdelta.co.uk/vmstjv5%20feb2005/decnet%20article%20vms%20tj%20v5%20feb2005.pdf and here: http://www.downloads.xdelta.co.uk/bootcamp%20jun2005/i220_decnet-plus_issue_1-1_10jun2004.pdf

Try to get to the "bootcamp" if you can: http://h71000.www7.hp.com/symposium/index.html?jumpid=symposium - based on all the DECnet questions of late in the ITRC I'll see if it's possible to re-run the DECnet session.

And yes, like Hoff, Bob, Hein, Steve Lieman et al - I also try to make a living by providing short-term consulting!

Good luck.
Cheers, Colin (www.xdelta.co.uk).

[Ken USA]

Oooops. Sorry did not that I had left this open. Our network expert was at our site to fix the problem.

Thank you all for your help and suggestions.