RE: NAT in the original Internet (RE: [wg-b] RE: [wg-c] IAB Technical Comment on the Unique)

[Harald Tveit Alvestrand <Harald@Alvestrand.no>]

At 22:41 19.12.99 -0800, Roeland M.J. Meyer wrote:

>In 1983, I was working at Retix, Santa Monica, building ISO protocol stacks
>and ethernet hardware (2 Mbps NICs). There were two outstanding issues
>preventing completion/finalization of the ISO protocols and they were very
>late. DOD became very impatient and decided to go with TCP/IP, as an interim
>measure, until the ISO got its act together (ISO bickering was a LOT like
>what the DNS warz sound like now <grin> [Note: within one year of TCP/IP
>implementation, the ISO "magically" came to agreement and finalized the
>spec, funny how that happened<grin>]).
>
>However, there were a number of nets already extant. The Internet was
>supposed to be the glue that bound them all together (therefore, the "inter"
>in "internet" ).

here's where our recollections diverge, I suppose.
In 1983, I wasn't working yet (showing my age!), but I was writing studies 
of the interconnection of OSI-based and TCP-based networks at school (NTH).
My conclusion, which was ultimately supported by the experience of othes, 
was that it was infeasible to try to concatenate two different service 
models; one had to adhere to either one, the other, or the minimum subset 
of the two. If one adhered to one, one would regard the other as a subnet.

The Internet was designed to ride on top of the others, not to interconnect 
them.

>  One of them was the French national network, which Retix
>was working on, with Olivetti, at the time. There were many others,
>including the uucp network, various IBM SNA networks, ARCnet, Netware
>systems, CompuServ, Time<something-or-other>, et al. All of them had
>different addressing schemes. The gateways for all of these required a form
>of NAT to work.

Application layer gateways, not NATs. You can't interconnect different 
service models, you can just gateway some subset of them. (Like telnetting 
into an X.25 PAD to connect to a terminal on an IBM mainframe - ouch!)
(In 1985, I worked on interconnecting document handling systems on SNA 
through an X.25 network to Norsk Data minicomputers. There was no doubt 
which was the subnet then....)

>  This is the basis of my statement. Interfacing to foriegn
>networks was the norm, at that time, and an intrinsic capability of the
>Internet.

Not of the IP-based network that became the Internet. The gateways were 
always at the edges.
And they were not as simple as NATs.

                        Harald

[I'm on the WG-C list now, so no need to forward my mail....]

--
Harald Tveit Alvestrand, EDB Maxware, Norway
Harald.Alvestrand@edb.maxware.no