[comp.protocols.tcp-ip] Many logical nets on a single physical net

fitz@wang.com (Tom Fitzgerald) (02/24/90)

Is it possible to have a single Ethernet with multiple IP network addresses
on it, and get the separate IP nets talking to each other?  Especially, is
it possible to do this cheaply?

We've got a number of separate Ethernets, with separate IP addresses, that
we want to connect together.  In the long run, as traffic rises (and
equipment becomes available) we'll be putting in proper gateways where
they look necessary.  But we want to get the machines on the various nets
talking to each other ASAP.

We don't want to renumber all the IP addresses, since we'd just have to
change them back when we started isolating the various nets with gateways.
And it may be some time before we can get all the gateways in place.

The ideal solution from our point of view would be a "gateway" with a single
interface and many IP addresses for it.  It would accept all inter-net
packets from all logical nets, and send them back out again, as-is, on the
same wire, to the right final destination.  Are there any gateways or
routers that are capable of this?  All the ones I have specs on assume that
interfaces map one-to-one with logical nets.

If there aren't any machines like this, what's the least we could do to get
this working?  Any and all info appreciated.

---
Tom Fitzgerald   fitz@wang.com
Wang Labs        ...!uunet!wang!fitz
Lowell MA, USA   1-508-967-5278

kwe@buit13.bu.edu (Kent England) (02/27/90)

In article <1990Feb23.214634.8645@wang.com> fitz@wang.com
 (Tom Fitzgerald) writes:
>Is it possible to have a single Ethernet with multiple IP network addresses
>on it, and get the separate IP nets talking to each other?  Especially, is
>it possible to do this cheaply?
>

	You need to be able to tell your hosts to ARP for all
addresses.  In other words, you need to make them think that every
other host is reachable directly.  The net part is zero length; the
entire address is "local part".  This gets you what you want without
the extra hop of a single-interface gateway.  If you add a real
gateway somewhere at some future time, and if it supports proxy ARP
correctly, this will work thru a transition from bridged/repeatered
LANs to properly internetted LANs, until such time as you can come
back to all your hosts and set a new subnet/net mask and stop using
proxy ARP if you choose.

	However, since there is a "built-in" net mask defining
net/local for class A, B, C, etc, this "zero-length" network mask may
not work in specific products, depending on how subnet masks were
implemented.  In fact, it might not be RFC legal.  (Must check that
Host Requirements RFC again.)  I don't know of anyone offhand who is
using proxy ARP this aggressively.  Most people just use it to handle
backward compatibility for the subnet problem between 4.2 and 4.3 BSD.
So their hosts are still using the "built-in" class-based mask or are
using subnet masks as an "extension", rather than "replacement" of the
built-in mask.  (Maybe you can set the subnet mask to be a negative
number? :-)

	But this does solve the dynamic gateway discovery problem
rather neatly.  That's it!  Let's do away with net/local altogether
and have hosts ask for pathways to every IP address they want.  Does
that sound like an advanced end-system protocol?  :-)

	Kent England, Boston University

bob@MorningStar.Com (Bob Sutterfield) (02/27/90)

You might try cobbling something up with Proxy ARP to fill in the ARP
tables when IP would rather route than ARP.