Re: Routing over Demand Circuits

Fred Baker <fbaker@acc.com> Thu, 19 August 1993 17:00 UTC

Received: from ietf.nri.reston.va.us by IETF.CNRI.Reston.VA.US id aa06086; 19 Aug 93 13:00 EDT
Received: from CNRI.RESTON.VA.US by IETF.CNRI.Reston.VA.US id aa06082; 19 Aug 93 13:00 EDT
Received: from atlas.xylogics.com by CNRI.Reston.VA.US id aa17017; 19 Aug 93 13:00 EDT
Received: by atlas.xylogics.com id AA30995 (5.65c/UK-2.1-930726); Thu, 19 Aug 1993 12:59:09 -0400
Received: from FENNEL.ACC.COM by atlas.xylogics.com with SMTP id AA27873 (5.65c/UK-2.1-930726); Thu, 19 Aug 1993 12:59:00 -0400
Received: from [129.192.64.5] (COAL.ACC.COM) by fennel.acc.com (4.1/SMI-4.1) id AA05768; Thu, 19 Aug 93 09:58:08 PDT
Message-Id: <9308191658.AA05768@fennel.acc.com>
Date: Thu, 19 Aug 1993 09:57:55 -0800
To: Gerry Meyer <gerry@spider.co.uk>
Sender: ietf-archive-request@IETF.CNRI.Reston.VA.US
From: Fred Baker <fbaker@acc.com>
X-Sender: fbaker@fennel.acc.com
Subject: Re: Routing over Demand Circuits
Cc: ietf-rip@xylogics.com

>You favoured TCP at Amsterdam.  Have you had any further thoughts on the
>'technical' merits?

You probed me privately, but I'm taking the liberty of responding publicly.
It seems to me that the points under consideration are:

    With or without TCP, I need something in the various RIP/SAP equivalents
    to change the way route entries are aged, I need at least a single backup
    route, and I need triggered updates. What we're arguing about is:

    A)  whether retransmissions and ACKs of the triggered updates go back
        to the routing protocol or an underlying distribution mechanism

    B)  whether it's harder to add a PDU to each of several protocols or to
        rehost them each on (a multiplexing layer on?) TCP
 
    C)  whether it's reasonable to require the assignment of IP addresses
        in, say, an AppleTalk-only network for the purpose of distributing
        Appletalk routing information. (Even if the point to point links are
        unnumbered, SOME interface in the device has to have an IP address in
        order to identify the device and satisfy TCP's addressing requirements.)

The argument for TCP is the commonality of the retransmission and
acknowledgement mechanism (A). The other two issues argue more towards your
proposal.

I would have to describe myself as leaning more than supporting, but at
this point I'm leaning in your proposal's direction, due to the network
management impact of (C). We find that many corporate networks have no IP
in them whatsoever except for the purpose of managing the routers, which
seems anachronistic.

BTW, regarding the question I raised the other day, I answered my own
question: if we assume that the data link protocols have a negotiation
mechanism like PPP's, when I go to forward a datagram to C via A, A will
refuse to bring up the datalink procedure for protocol foo. At that point,
I know that I have to go via B.
=============================================================================
Don't blame ACC; they think I'm nuts too!