Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!uwm.edu!bionet!agate!ucbvax!ucsd!hub.ucsb.edu!spectrum.CMC.COM!lars
From: lars@spectrum.CMC.COM (Lars Poulsen)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Routers, IP networks, and wide-area networks
Message-ID: <1991Mar26.192447.3009@spectrum.CMC.COM>
Date: 26 Mar 91 19:24:47 GMT
References: <9103211614.AA22207@leo.md.interlink.com>
Organization: Rockwell CMC, Santa Barbara, CA
Lines: 62

In article <9103211614.AA22207@leo.md.interlink.com>
   cam@md.interlink.com (Chris Markle) writes:
>When IP routers are connected to each other via X.25 networks, is the X.25
>network treated as a single IP network/subnetwork, or is each "point-to
>point" link (ie. X.25 virtual circuit) treated as a separate network/sub-
>network? (Or are the "links" not treated as networks at all?)

The IETF/IPLPDN group has just been discussing analogous stuff regarding
Frame Relay, so this is a resonably warm topic.

When IP-over-X.25 was first defined, and implemented for CSNET, network
number 14 was reserved as the IP representation of the global X.25
internet.  So the whole IP world was ONE IP subnet.
(Only the CSNET subscribers ever made it onto the list, and there was
no simple mechanism for other sites to register, or to perform address
translation lookups, so it fizzled.)

When MILNET/ARPAnet implemented X.25 host-to-IMP protocol, the IP
network(s) already existed, and X.25 addressing was a thin overlay.
So MILNET was one IP subnet, and ARPAnet was another IP subnet.

When SUN implemented their commercial IP/X25 encapsulation product, they
chose to view the X.25 interface as a collection of point-to-point
connections. One consequence of this was that the host had a different
IP address for each remote host that it connected to.

ACC's commercial X.25 products follow the "one subnet" model. So do
CMC's.

>For example, assume that I have an X.25 network with three routers connected
>to it, R-a, R-b, and R-c; each router knows about the other two. Each router 
>is attached to a single Ethernet; for simplicity, each Ethernet is a separate 
>IP network - R-a is attached to 192.1.1, R-b to 192.1.2, and R-c to 192.1.3 
>(class C networks all).
>
>Would the X.25 network be assigned a single network number, say 192.1.4, and
>each router would have an X.25-side IP address of 192.1.4.x? Or would each
>"link" (VC) between each router have network numbers assigned to it (eg.
>R-a <-> R-b is 192.1.4, R-a <-> R-c is 192.1.5, and R-b <-> R-c is 192.1.6)?

Both models are "common". They do not interoperate well.

The "one subnet" is easier to manage. The "point-to-point" model allows
certain configurations that are difficult to implement with the "one
subnet" model.

>(If each link gets an IP network/subnetwork address, it seems like in a big 
>X.25 or other wide-area network many addresses will be chewed up just on these
>point-to-point links. Even subnetting results in many subnets addresses being
>consumed. Like Arsenio says, "hmm...")

Actually, Berkeley-based IP routers know that point-to-point links have
one IP address at each end, and are not necessarily "networks". It is
still a pain to handle.

>Chris Markle
>Interlink Computer Sciences

Say Hi to the gang for me !!
-- 
/ Lars Poulsen, SMTS Software Engineer
  CMC Rockwell  lars@CMC.COM