Re: [BEHAVE] proprietary implementation v.s standardisedprotocols//re: draft-xu-behave-nat-state-sync-00

[Simon Perreault <[e-mail]>]

Andrew Sullivan wrote, on 2009-12-01 14:40:
> On Tue, Dec 01, 2009 at 07:26:29PM +0000, Dave Thaler wrote:
>> So yes NAT64 should try (to the extent that is practical) 
>> to accommodate apps that don't pay attention to TTL, since that's
>> 99.99% of all applications.
> 
> Either that, or we should get to work on changing the APIs.  Since I
> predict heat death of the universe first, I agree with Dave.

No, let's change the APIs to make NAT64 work!

Just kidding... we all agree on that. But what was the point again?

The point was that you cannot fail over clients from one Pref64::/n to another
because applications will remember addresses. This is true, to some extent. If
it was 100% true then you wouldn't be able to change anything once you publish a
DNS record. In reality, hosts are rebooted, applications are restarted, and web
pages are reloaded. So you have some wiggle room.

Failing over to another Pref64::/n is an unlikely event. It means that something
*really bad* happened. (See my other email about the two types of breakage.) Any
partial fix is better than just shutting of the Internet to your customers. If
it means they have to close Outlook and restart it, so be it.

Simon
-- 
DNS64 open-source   --> http://ecdysis.viagenie.ca
STUN/TURN server    --> http://numb.viagenie.ca
vCard 4.0           --> http://www.vcarddav.org