Isn't that closer to Proxy ARP? RFC 1027 credits RFC 925.
At CERN, we used an unpublished ad hoc NAT mechanism in about 1980
to interconnnect two copies of a homebrew network with absurdly
small addresses. The DECnet Phase IV 'hidden areas' mechanism was
also a widely used NAT-like hack in the 1980s.
Of course I have no idea of the details of whatever CERN did back in the 80's,
but the bit about hidden areas didn't gibe with my memory so I looked it up
to be sure...
DECnet hidden areas weren't really NAT-like at all. They were simply a set of
addresses in one group of areas that weren't visible across one or more level 1
routers to another group of areas. In order to get to a hidden area in another
group, you had to use explicit multi-hop routing, e.g., assuning STAR is the
gateway between groups and you want to reach a system called XDELTA in the
other group, you have to say STAR::XDELTA:: instead of just XDELTA::. (I
believe both STAR and XDELTA were actual system names on Digital's large Phase
IV network, chosen because "star" was the code name for the original VAX-11/780
and "xdelta" was the name of the kernel debugger.)
There are plenty of parallels to this in email, including UUCP routing (a!b!c),
percent-hack routing (c%b@a), source routing (@a:c@b), and mixed routing
(b!c@a). (Note that this covers four of the six available permutations - I'd be
curious to know if anyone has an example of either b-a-c or c-a-b order being
used anywhere.)
In fact the same term was used to refer to this trick in both DECnet Phase
IV and email: Poor Man's Routing (PMR).
NAT would be a lot less popular than it is if explicit routing was needed on
all the end systems to make it work. But of course it doesn't work like that.
Ned