Re: Multihoming Issues

Fred Baker <fred@cisco.com> Fri, 30 August 2002 19:46 UTC

Received: from loki.ietf.org (loki [10.27.2.29]) by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA02837; Fri, 30 Aug 2002 15:46:25 -0400 (EDT)
Received: (from adm@localhost) by loki.ietf.org (8.9.1b+Sun/8.9.1) id PAA08645 for ietf-outbound.09@loki.ietf.org; Fri, 30 Aug 2002 15:34:01 -0400 (EDT)
Received: from ietf.org (odin.ietf.org [10.27.2.28]) by loki.ietf.org (8.9.1b+Sun/8.9.1) with ESMTP id PAA08610 for <ietf-mainout@loki.ietf.org>; Fri, 30 Aug 2002 15:30:11 -0400 (EDT)
Received: by ietf.org (8.9.1a/8.9.1a) id PAA02124 for ietf-mainout@loki.ietf.org; Fri, 30 Aug 2002 15:28:39 -0400 (EDT)
X-Authentication-Warning: ietf.org: majordom set sender to owner-ietf@ietf.org using -f
Received: from sj-msg-core-1.cisco.com (sj-msg-core-1.cisco.com [171.71.163.11]) by ietf.org (8.9.1a/8.9.1a) with ESMTP id PAA02101; Fri, 30 Aug 2002 15:28:13 -0400 (EDT)
Received: from FRED-W2K6.cisco.com (stealth-10-32-253-235.cisco.com [10.32.253.235]) by sj-msg-core-1.cisco.com (8.12.2/8.12.2) with SMTP id g7UJTDKC004179; Fri, 30 Aug 2002 12:29:13 -0700 (PDT)
Message-Id: <5.1.1.6.2.20020830095509.07527190@mira-sjcm-4.cisco.com>
X-Sender: fred@mira-sjcm-4.cisco.com
X-Mailer: QUALCOMM Windows Eudora Version 5.1.1
Date: Fri, 30 Aug 2002 10:16:40 -0700
To: Sister Sibling <ccs522g9@yahoo.com>
From: Fred Baker <fred@cisco.com>
Subject: Re: Multihoming Issues
Cc: ietf@ietf.org, iesg@ietf.org
In-Reply-To: <20020830021621.91892.qmail@web12401.mail.yahoo.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format="flowed"
Sender: owner-ietf@ietf.org
Precedence: bulk
X-Loop: ietf@ietf.org

At 07:16 PM 8/29/2002 -0700, Sister Sibling wrote:
>The IETF is recommending that the DNS mechanisms to support IPv6 stay 
>essentially the same as those already in use with IPv4 today. To our 
>opinion, in the realm of multi-homed networks, the techniques used in IPv4 
>can't all be applied since they have scaling problems. Specifically, if 
>the same prefix is advertised by multiple ISPs, the routing tables will 
>grow as a function of the number of multihomed sites.

It seems like routing tables and DNS are separate discussions - one is 
about routes and prefixes, and one is about names. Since I don't know what 
your DNS concern is, I'll leave it for the moment.

On the routing problem, you have a point, but it is one that should be 
solvable. There is, as you know, nothing magic about IPv6 prefixes with 
respect to this, but we do have the opportunity to issue an entirely new 
set of prefixes with an entirely different mindset. Whereas today the RIRs 
hand out relatively short prefixes with a view to forcing the use of NATs 
in edge networks, their plan is to give each ISP a prefix large enough to 
give a /48 to each of its customers. This gives each customer the option of 
defining 65K subnets or some amount of structure within itself - a lot like 
handing every IPv4 customer a Class A Address and having him use the least 
significant 8 bits for a host number.

In such a scenario, in your favorite location in the network, there should be:

  - one prefix for each ISP in the world
  - one prefix for each POP or campus in your network
  - one prefix for each LAN in your POP or Campus
  - additional prefixes that you decide to carry for your own reasons (eg, 
policy)

The "additional prefixes" that an ISP carries might, for example, include 
/48 prefixes from customers who got their address from another ISP - 
multihomed addresses. I expect, however, that while the major ISPs would 
want to advertise these to their customers, they would find it in their own 
best interest to not advertise those to other major ISPs, and to not accept 
them from other major ISPs if they are advertised back. This is comparable 
to the existing reportedly-common policy of advertising and accepting only 
prefixes comparable in size to the RIR's allocation units.