shim6 (was Re: IPv6 news)

Fri Oct 14 22:07:06 UTC 2005

On Fri, 14 Oct 2005, Paul Vixie wrote:

> david.conrad at nominum.com (David Conrad) writes:
>
> (shouldn't that be drc at iana.org now?)
>
>> If my impression is correct, then my feeling is that something else
>> is required.  I am somewhat skeptical that shim6 will be implemented
>> in any near term timeframe and it will take a very long time for
>> existing v6 stacks to be upgraded to support shim6.  What I suspect
>> will be required is real _site_ multihoming.  Something that will
>> take existing v6 customer sites and allow them to be multi-homed
>> without modification to each and every v6 stack within the site.
>
> if all you've got is a hammer, every problem looks like a nail.  so, the
> above problem statement looked like a dns issue to me, and to some other
> folks, and thus was born A6.  had ietf killed AAAA back when there were
> effectively zero ipv6 hosts on the 'net, and paid the apparently-high A6
> complexity penalty, we'd be talking about something else by now.  as it
> is, the shim6 complexity penalty is even higher, and i don't think we'll
> ever get to stop talking about this problem.

Paul,

I was big fan of A6 design and very sorry to see it discarded by IETF.
But I do not believe it would have solved quite the same problem as what
shim6 is trying to do (I'm not saying that it could not solve it with
additional ipv6 extensions...). A6 was particularly good for solving
problems of renumbering and allowing site to have same local ip address
configuration for multiple ISP connections which can allow easy load
balancing through DNS. But once connection is established it would
be normal end-node<->end-node IPv6 connection with full 128bit ip6
addresses used for each side. That means that if connection goes
down it would not be able to automatically do a fail-over (nor would dns 
load-balancing similar to current one with listing multiple A addresses
do failover to use only active and working ipv6 connection).

What I do think is that A6 and shim6-like design would have been a good
compliment to each other. In this case the end-site network address
(i.e. part above /48) would have been the same and is the one configured 
on switches and servers and listed as host address. The domain reference
(i.e. common to multiple hosts) would point to the ISP networks A6 network 
addresses (i.e. part below /48) and is the only thing one changes when 
moving from one ISP to another (no renumbering!) or when connecting 
to 2nd ISP. If multi6 is used then additional special reference is made
to multi6 network address (special reserved network block non-routable 
block) and multi6 aware clients would use that as common ip6 address
reference for upper layers (i.e. in TCP and UDP) and let very simple 
multi6 layer keep track of local host's resolution of this address to 
real A6 networks.

-- 
William Leibzon
Elan Networks
william at elan.net