Path: utzoo!utgpu!news-server.csri.toronto.edu!bonnie.concordia.ca!thunder.mcrcim.mcgill.edu!snorkelwacker.mit.edu!apple!usc!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!lll-winken!prang!ejmmips.NOC.Vitalink.COM!ejm
From: ejm@ejmmips.NOC.Vitalink.COM (Erik J. Murrey)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Redirects, and multiple subnets on a cable
Message-ID: <31@prang.TEST.Vitalink.COM>
Date: 21 Jan 91 23:30:41 GMT
References: <6490@munnari.oz.au>
Sender: usenet@prang.TEST.Vitalink.COM
Reply-To: ejm@ejmmips.NOC.Vitalink.COM (Erik J. Murrey)
Organization: Vitalink Communications
Lines: 42
Nntp-Posting-Host: ejmmips.noc.vitalink.com


In article <6490@munnari.oz.au>, kre@cs.mu.OZ.AU (Robert Elz) writes:
> 	In there you say (wrt two subnets on one cable) ...
> 
> 	   In fact, this is not such a bad solution,
> 	   because assuming that the gateway is capable of recognizing multiple
> 	   subnet numbers on the same subnet, the gateway will simply send the
> 	   host an ICMP Redirect [4], and subsequent packets will go directly to
> 	   the host [1].
> 
> 	I don't think that can be true can it?  That would require the ICMP
> 	redirect to contain an ethernet address.  The sending host is in no
> 	doubt of the destination's IP address, sending a redirect that contains
> 	that address can do no more than confuse it, if its routing table has
> 	it believe that to reach that address it must route through a gateway.
> 
> 	Oh - do you mean send the host a redirect, containing its own address
> 	as the gateway?  I guess that might work, assuming that the host's
> 	software understands the BSD type convention "if I am the gateway,
> 	I send directly out on my ethernet", if not, almost anything might
> 	happen.
> 
> 	If that's it, I think I'd be more explicit about it.
> 


Actually, using two subnets on a cable is very common when more than
one link-layer protocol is being used.  (i.e. Ethernet v2 and
802.3/SNAP)  In this case, an ICMP isn't even practical since we don't
have a way to specify link-layer information in an ICMP message.

In cases where the subnets share the same frame type, it is still
impractical to send ICMP's since the "offending" host will know
nothing about host specified in the ICMP redirect.  Most IP
implementations based on the 4.3Tahoe stack only allow one IP address
per interface.  Maybe the interface handling code could be extended to
support multiple addresses per protocol?

---
Erik J. Murrey
Vitalink Communications NOC
ejm@NOC.Vitalink.COM	...!uunet!NOC.Vitalink.COM!ejm