DECNET how to change adjacent node

[Robert Gezelter]

Juan,

I concur with Volker on the NOD* vs SYS*. The information from the network having the problem is needed (e.g., NOD*).

Also, what does the Ethernet configuration look like beyond the OpenVMS systems? Has the backbone configuration been changed?

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

[Hoff]

DECnet is not IP, and DECnet routing does not work like IP gateway routing.

There is typically no error and no issue and no need to mess with DECnet (level one) routing on an Ethernet, save for cases where there's a woefully underpowered or overloaded server acting as a DECnet designated router, and that's not common these days, and would be unexpected with a half-dozen hosts on an Ethernet.

If you want to force this configuration (and constrain the options DECnet has here) (and there can be reasons to implement this), then the usual approach is to migrate to DECnet area routing (level two routing), and separate the functional or geographic differences into unique DECnet areas.

If your network is geographically dispersed (as is implied by that flaky network segment), then the usual solution is enabling the hosts on either end of the point-to-point connection link as DECnet area routers. (If this is a VLAN connection involving switches or such, then those DECnet area routers might not be directly connected to the point-to-point links.)

Irrespective of routing, DECnet end-node hosts can and do spot adjacencies on an Ethernet, and no DECnet routers are required when all hosts involved are on the same (logical or physical) Ethernet; end-nodes can communicate directly.

(I don't remember off-hand if the DECnet end-nodes can spot each other across areas. That configuration isn't particularly common, in my experience; by the time folks are rolling out area routing, there are usually several routers active on each network segment and one or more area routers will be active, so the end-nodes will have connectivity.)

[Hoff]

ps: Try to fix that flaky or overloaded network segment.

Flaky networks play havoc with DECnet and with IP. Back-offs and retransmits can cause cascading (and nasty) communications failure modes.

And they expose application bugs.

And FWIW, many applications aren't particularly good at recognizing and recovering from flaky networks and from network disconnections, meaning that modern (mobile) network communications can cause some serious disruptions within these applications.

[Volker Halle]

Juan,

if the SYS11x nodes are the production environment and you 'have problems' in the production environment, then what are these problems ?

The 5 SYS11x nodes seem to see each other on the LAN as shown by your attached .TXT file. Can you do SET HOST between all of them ?

Volker.

[Juan M. Lazaro]

Hi and thanks to all!

Yes I have switches between the hosts, the problem is in SYS11X configuration, forgot NODXX.

I can acces via SET HOST between all nodes in the area, the problem is than between SYS11D and SYS11B, I have frecuent connection timeouts in the EIA-0 circuit, and I supose it is because first it goes to SYS11C machine (adjacent SYS11D node, and in PLACE B), and then it returns to SYS11B, in PLACE A, I'm not sure how DECNET routing works, for those I want to put like adjacent node in SYS11D the node SYS11B.

Many thanks to all.

Juan

[Robert Gezelter]

Juan,

Ok. I strongly suggest doing a NCP SHOW CIR EIA-0 COUNT on the nodes that are having a problem. I would also be strongly inclined to use WireShark to look at precisely what is transiting the actual backbone.

As Hoff noted, LAN problems can have all manner of strange side effects. The error counters are extremely useful, as is the trace of what is actually happening on the network segment.

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

[Juan M. Lazaro]

Hi Robert,

I must put SYS11D out of production mode and now is in other network (TEST), I was doing some tests changing the executor type to nonrouting IV, but the only test I can do is in production with the application running on it...

But when I have the problem, with the command

SHOW COUNT NODE SYS11B

I have 14 response timeouts, but I don't have the counters from the circuit, and the client don't let me put the machine in production mode.

Thanks

Juan

[Volker Halle · Accepted solution]

Juan,

let's start with some basics on how DECnet routing works:

- each end node sends an Endnode Hello multicast message to all routers every Hello Timer seconds

- each routing node sends an Ethernet Router Hello multicast messages to all DECnet nodes every Hello Timer seconds

That's how the routers learn about the end node and the end nodes learn about the routers and the routers learn about each other.

If an application on a node wants to connect to another application on another node, it sends the first packet to the designated router (typically the router with the highest DECnet address on the local LAN), who then forwards the packet to the destination node. If the destination node is on the SAME LAN, it sets the 'intra ethernet bit' and future packet exchange happens directly between the 2 nodes on the same LAN.

Reponse timeouts are NODE counters and count how often a packet sent via a logical link to another node is not being acknowledged in time. This is typically an indication of DECnet packets getting 'lost' in the network path between the communicating nodes.

You should first look at the DECnet LINE counters and at all physical and datalink layer counters of all network components between the 2 systems involved.

If the node has now been removed from the production network, you can still at least test it's ability to communicate with other nodes in the test network. Using the DTSEND utility, you can test DECnet packet exchange between any 2 DECnet nodes and generate a lot of traffic to test the network components.

Volker.

[Juan M. Lazaro]

Hi Volker,

Many thanks to your time expent in that, and about the explanation on how DECnet runs, it is more clear to me now...

I'm going to look DTSEND utility and try to reproduce the error in test and let you know the results.

Thanks to all the people who help me in that.

Cheers,

Juan

[Juan M. Lazaro]

Perfect!!! great utility DTSEND... I was testing the DECNET network, at the end, I saw than have an elevate number of Response timeouts, I have decide to change the network card, and pass the tests again, and it works correctly, I haven´t hade more response timeouts, while running the test, in the other node I have a

MONITOR DECNET /ALL/INTER=1

running and I can see how some times there is no traffic while the DTSEND is running, and after that, looking the counters of the node I'm sending the traffic can see one respones timeout plus than before...

Thanks a lot to all for your help, I couldn't understand DECNET network without your explanations.

Have a great day,

Juan