[HECnet] Multinet peerings...?

[Johnny Billquist]

On 2016-01-15 03:25, Brian Hechinger wrote:
>
>> On Jan 14, 2016, at 8:03 PM, <Paul_Koning at Dell.com> <Paul_Koning at Dell.com> wrote:
>>
>>
>>> On Jan 14, 2016, at 4:40 PM, hvlems at zonnet.nl wrote:
>>>
>>> ...
>>>> Can the bridge program detect whether there are area routers for the dame area at both ends and favor the local one, possibly block advertising of the remote area router?
>>>
>>> I tried blocking traffic from a node in one area from getting to another
>>> area, with the exception of packets from area routers.
>>> Unfortunately, it does not work. DECnet can be clever about local
>>> ethernet connectivity. If you are on the same ethernet segment, nodes
>>> can communicate directly with other nodes on the same ethernet segment,
>>> even if they are endnodes, and this exen extends to nodes on different
>>> areas. So such filtering in the bridge cause communication to fail for
>>> endnodes on the ethernet segment, when the destination is on the same
>>> ethernet, even if in a different area.
>>
>> DECnet expects a "transitive Ethernet" -- if A can talk to B and B can talk to C, A must be able to talk to C.  That's actually a common assumption, other network protocols do the same.  DECnet is a bit unusual in that it explicitly verifies this property, at least for routers -- that's why router hellos have the router list in them.  We put that in because we had run into some defective Ethernets that were non-transitive, causing very strange misbehavior until this protocol mechanism was added.
>>
>> End nodes have an on-Ethernet cache: if X talks to Y and both are on the same Ethernet, they will do it directly.  From the first packet if there are no routers; after the initial round-trip if there are.  If you create a non-transitive Ethernet -- which is what filtering does -- this will fail.  There is no workaround.  If you don't want all the nodes on an Ethernet to have direct communication, the only solution is to split it into two separate Ethernets, interconnected by a router (not a bridge).
>
> And this is what I was (and always have been) thinking. Make “shorter” ethernet segments that are less geographically diverse. Put routers between them. That should solve most problems, no?

Yes. This is actually just the traditional way networks are designed. 
Ethernet is a LAN - as in local. It's not designed for long haul 
connections. It only works because the internet today have pretty 
amazing capacity compared to the 80s.

But to get a more traditional topology, we need the routers in between 
somehow - and the WAN links.
The problem with that have been that much DECnet gear only supports 
various links that best would be described as arcane by todays 
standards, in addition to the ethernet. How many use X.25 nowadays? Or 
synchronous serial lines?

Well, if you were running a Cisco box, you could tunnel DECnet over GRE. 
Not everyone have one of those. The other option more "generally" 
available was VMS machines running Multinet, as that supported DECnet 
links carried over IP.

One option that is slowly becoming more and more plausible are specific 
routers, such as Pauls python router, and Rob Jarrats implementation.

I can now happily add another. I think I found the annoying bug in my 
Multinet-compatible DECnet driver for RSX, so now RSX can also haul long 
links over point-to-point instead of having to rely on the bridge.

First link, between MIM and LEGATO is up, and looking stable.
More coming, I hope. Steve, how about SG1? ;-)

Peter, would you like a link or two between somewhere at your end and 
MIM as well? Maybe less useful, since we already have DIMMA that does 
the same.

	Johnny

-- 
Johnny Billquist                  || "I'm on a bus
                                   ||  on a psychedelic trip
email: bqt at softjar.se             ||  Reading murder books
pdp is alive!                     ||  tryin' to stay hip" - B. Idol