Path: utzoo!attcan!uunet!husc6!think!ames!eos!labrea!agate!ucbvax!csp-a.prime.COM!MIKEC
From: MIKEC@csp-a.prime.COM
Newsgroups: comp.protocols.tcp-ip
Subject: (none)
Message-ID: <8805141153.AA10531@ucbvax.Berkeley.EDU>
Date: 13 May 88 19:59:30 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 28

To:       (smb@ulysses.att.com)
Cc:       (tcp-ip@sri-nic.arpa)
From:     Michael A. Curtis (mikec@csp-a.prime.com)
Date:     13 May 88  3:40 PM
Subject:  Re: Smart vs. dumb host routing

Steve,

     I think you've missed the whole point of the issue.  That is the exact
intent of ICMP in this case.  Once the host delivers the packet, its job is
over and if it is sent to the wrong default gateway for the destination
network chosen, then an ICMP (*HOST*) redirect should be issued telling the
host which gateway is the correct one to deliver future packets for that
network to.

     Further, once the gateway has the packet, ICMP is out of the picture as
all gateways should have up-to-date knowledge as to what the topology
currently looks like via a routing protocol such as RIP.  There should be
absolutely no indecision as to whether it should deliver the packet to G'
or G'', etc. and no need for ICMP redirects to be flying around.  Also, EGP
probably has no bearing at this point as the discussion has been focusing on
the problems/advantages of subnets with different size masks, thus the traffic
would be internal to the AS and EGP doesn't know about subnets anyway!

     As a separate discusion, we can talk about the merits of having a
different subnet mask on a per interface basis in a gateway as discussed
in RFC1009.  To my knowledge, no one is implementing this.  Does anyone know
of any plans?