[prev in list] [next in list] [prev in thread] [next in thread] 

List:       ipng
Subject:    (IPng 4982) Re: change IPv6 DNS records....
From:       huitema () bellcore ! com (Christian Huitema)
Date:       1997-11-26 14:44:23
[Download RAW message or body]

On Nov 25,  5:51pm, Mike O'Dell wrote:
> Subject: (IPng 4978) Re: change IPv6 DNS records....
>
> No, it should not move to Draft with known serious scaling problems.
>
> that is one of the things V6 is supposed to address

Let's debate this in DC.  What the draft proposes, in the new AAAA part,
is the best possible support of network renumbering -- I can actually
demonstrate that we cannot do better than meeting the "store a prefix
exactly once" property.  But it also proposes an automic support of
multihoming, a point that seem to be lost when I hear arguments such as
"but my ISP is multihomed."

Let's be clear.  Short of generalized address rewriting inside the
network, there are three ways to support ISP multihoming:

1) allocate a single routing prefix to every multihomed ISP, regardless of
   its size.

2) implement geographic addressing.

3) allocate several routing prefixes to multihomed ISP, and derive from
   these prefixes an equal number of addresses for their customers.

Solution number 1 is, by and large, what we are doing with IPv4.  We have
experience there, and the experience is ugly: the solution simply does not
scale.  Think of it: my coop board is installinga router to serve the
building's resident. It thus become, de facto, a tiny ISP.  And it will
very probably multi-home.  Solution 1, without a limit to the ISP size,
would lead to "one entry in the global routing table per appartment
building."  Do you really like it?

Solution 2 requires a massive investment in geographical exchanges, which
will turn in de facto local monopolies.  This may or may not happen in the
future.  I would not bet on it, let alone mandate it.

Solution 3 is actually supported by IPv6 address configuration mechanisms,
which can routinely assign several prefixes to a link and generate several
addresses per interface.  It places the burden where it should be, that is
on the fringes of the network, in the stations.  This is not an infinite
burden -- merely handling a list of addresses.  What the DNS draft
proposes is to complement a three piece architecture comprised of:

* address configuration, either automatic or through DHCP,

* router renumbering,

* support of automatic multihoming in the DNS.

Now, supporting this is not equivalent to mandating it.  ISP, at some
point, may grow sufficiently and graduate to the "top" division; they will
then get a top level address.  And some governments may well mandate the
use of monopolistic local exchanges.  Who knows?  the soviet mode of
organizing society may stage a come back :-).  But we are not in the
business of policing the allocation of addresses, let alone mandating
societal changes (specially that one).  We are in the business of
engineering IPv6.

My position has not changed since August.  I believe that we should
implement the proposed change.  I also believe that we should tweak the
proposal to make it "upward compatible", in order to avoid a flag day on
the 6Bone.

-- 
Christian Huitema
--------------------------------------------------------------------
IETF IPng Working Group Mailing List
IPng Home Page:                      http://playground.sun.com/ipng
FTP archive:                      ftp://playground.sun.com/pub/ipng
Direct all administrative requests to majordomo@sunroof.eng.sun.com
--------------------------------------------------------------------

[prev in list] [next in list] [prev in thread] [next in thread]