To:
dnsop@cafax.se
From:
Rob Austein <sra+dnsop@hactrn.net>
Date:
Wed, 20 Mar 2002 14:09:06 -0600
In-Reply-To:
<32650000.1016637507@slimsixten>
Sender:
owner-dnsop@cafax.se
User-Agent:
Wanderlust/2.8.1 (Something) SEMI/1.14.3 (Ushinoya) FLIM/1.14.3(Unebigoryòmae) APEL/10.3 Emacs/20.7 (i386--freebsd) MULE/4.0 (HANANOEN)
Subject:
Re: draft-ietf-dnsop-v6-name-space-fragmentation-01.txt
Might I suggest that, since this entire discussion is about a relatively minor point in Johan's draft (which, if you read carefully, you will notice Johan is not even suggesting we should do, he's just trying to show where one line of thinking might lead), it might make more sense to discuss the main content of the draft, to wit: what the bleep are we going to do about this fragmentation problem? For folks who didn't make it to the face-to-face WG meeting yesterday, here's the straw man I outlined towards the end of the meeting as a possible strategy. It'd be nice if we could do better than this, but that's what we need to discuss. This straw man does not address all of the issues in Johan's draft or presentation, but I think that part of Johan's point is that we are going to have to compromise on some things because this is a choice among various bad options. Anyway: Start out with all zones (v4 or v6 content) available via v4 transport. Use dual stack resolvers (or forwarders or whatever) in IPv6 land until IPv4 is dead. In latter stages, this probably means tunneling IPv4 addresses into IPv6-only clouds for this purpose. Eventually this will start to break down. Customers who are just in the v6 cloud (lack access to dual stack resolver for whatever reason) will start complaining about not being able to reach sites (will look like ordinary failures). V4-based servers will push their zones into v6 space as well, if they care, if not, perhaps we don't care about them. Servers that are just in v6 land will start popping up when v4 addresses get really hard to obtain. Clients stuck in v4 land will whine to their ISPs, who will install translation boxes or will upgrade the customer's software or whatever so that customer is now on v6. Eventually the number of people remaining in the v4-only cloud will not be worth supporting, and will be orphaned. This will suck, but it's what happens with every technology change. So we live with it. Johan and I expect to do some work on further detailing the problem. So please read the above as an attempt to provoke discussion, and perhaps as a dead-simple model against which others can be measured (that is: anything that's obviously even worse than this proposal probably need not be considered). --Rob