Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!elroy.jpl.nasa.gov!ames!cincsac.arc.nasa.gov!medin
From: medin@cincsac.arc.nasa.gov (Milo S. Medin)
Newsgroups: comp.unix.wizards
Subject: Re: laptop internet addresses
Message-ID: <1991May31.013956.6415@news.arc.nasa.gov>
Date: 31 May 91 01:39:56 GMT
References: <27035@adm.brl.mil> <1991May30.084135.25587@thunder.mcrcim.mcgill.edu>
Sender: usenet@news.arc.nasa.gov (USENET Administration)
Reply-To: medin@cincsac.arc.nasa.gov (Milo S. Medin)
Organization: NASA Science Internet Project Office
Lines: 67

In article <1991May30.084135.25587@thunder.mcrcim.mcgill.edu>, mouse@thunder.mcrcim.mcgill.edu (der Mouse) writes:
|> In article <27035@adm.brl.mil>, ketell@mercury (Gregory Ketell) writes:
.
.
.
|> 
|> > Why bother giving them multiple addresses at all.  Give each machine
|> > a single address, whatever network they hook up to they will still be
|> > reachable.  (ex.  You can reach me, I can reach you but we are not
|> > only on different subnetworks but across the nation.).
|> 
|> Only because the routing algorithms take advantage of the fact that at
|> each level of subnetting, all the way from the network number down to
|> the smallest subnet, each network is fully connected within itself.
|> Your proposal breaks this.
|>

Not quite.  OSPF was designed with variable length subnet in mind.  There is
no notion of class A, B, or C nets in OSPF.  Every net is passed with a mask.
Note this also obviates the need to have subnetted nets be contiguous.  You
can have pieces of a class B connected via a Class C in the middle.  This
works fine.  Since EGP and BGP do not pass subnet info around, this
functionality is limited to the insides of an OSPF system, but that's still
very useful.  All vendors shipping OSPF implementations support this kind
of variable length mask support.  So, you should be able to pull this off.
  
|> > Just assign each machine the IP address for their "Home Connection",
|> > ie if they are in your math subnetwork just assign them an address on
|> > that subnet.  When they hook up to another net they will operate on
|> > that net with their own address.
|> 
|> > Admittedly this will only work as long as both subnets can reach each
|> > other.
|> 
|> It will also be a major nightmare for the routing protocols.  RIP,
|> which I would guess is the commonest routing protocol within small
|> (campus-sized) networks, cannot deal with this.  Most UNIX boxes will
|> be unable to reach the machine when not on its home subnet, unless they
|> are told that the network isn't subnetted at all and then the gateways
|> made to do proxy arp - which opens up a whole 'nother can of worms.
|>

As I said, OSPF does not have this problem.  I expect RIP use to gradually
fade away until it's just used as an easy way for hosts to find routers, (ie
all routers just send RIP default) and not for much else.  In fact, if you
really take advantage of this, you could pull a laptop off the LAN interface,
and have it keep it's address when it dials into a router via SLIP or PPP.  When
on the LAN, the "most specific route" points to the LAN itself, and everything
is as it works now.  When dialing into a terminal server acting as a router
and speaking OSPF, the server orginates a host route to the target address.  
The host route is more specific and thus the network routes that way instead
of to the LAN (for just that host).  The router attached to the LAN can proxy-arp
so hosts on the LAN can talk to the remote laptop.  All of this is doable
with OSPF.  Now, you just need to have your terminal servers run OSPF (since
a well known vendor's server is built on top of BSD Unix, and since gated will
shortly support OSPF, this isn't as far fetched as it may sound), and viola!
The IETF actually does do good work from time to time.  OSPF is a draft
internet standard right now, and hopefully will go to full standard (like
SNMP) in 6-9 months.  Ask for it by name, accept no substitutes!

I'm a little bit biased here, since I worked on the design of OSPF, but I think
it will handle the case you are thinking about.
   
						Thanks,
						   Milo

PS Usual disclaimers apply...