Re: spending time on analysis [Re: draft-palet-v6ops-proto41-nat-03 as WG item]

[Tim Chown <tjc@ecs.soton.ac.uk>]

On Mon, Nov 17, 2003 at 12:45:06PM +0200, Pekka Savola wrote:
> 
> You can devise a use for every tool out there, and more besides.   Writing 
> such an applicability statement is trivial; a couple of years I made the 
> case myself in a paper I wrote.  But that doesn't mean those methods are 
> necessary or useful.

I agree some level of analysis is very useful, both for specifics of the
mechanisms, and how they interact with other mechanisms.   But also the
requirements on the whole architecture are important, e.g. Teredo and 6to4 
both like talking to other Teredo or 6to4 systems respectively and would 
require a heavy investment in relays if they are to communicate with native 
IPv6 systems.

I agree that ideally the WG needs to decide if certain tools are required
from the analysis.   But that result will depend on how the scenario 
document is phrased.  For example, a scenario of dual-stack hosts in an 
IPv6-only infrastructure could quite reasonably lead to a DSTM-like solution.
Likewise, the device-behind-a-NAT scenario could lead to a Teredo-like 
solution.  The question then becomes what are valid scenarios, as any scenario 
can lead to a certain best (or only) solution.   And how do we judge those
solutions in the big picture (e.g. the requirement of wide-spread deployment 
of Teredo or 6to4 relays in ISP networks?).

> The main point of v6ops (AFAICS) is to show the folks how to start
> deploying IPv6 in the most common scenarios.  It'd both act as a guide how
> to go on with it ("look, you can do X, you don't need Y"), and as a way to
> identify required transition mechanisms which will be necessary.

I think the basic transition mechanism doc is fine, and we have basic
tools for "structured" deployment - native, dual-stack, and (though loosely 
specified) tunnel brokers.   It seems some of the reluctance to adopt certain 
tools has a tension between the "perfect transition" and the "reality now", 
as was reflected in the IAB/IESG Minneapolis presentation - do we define 
standards for the Internet, or standardise what is on the Internet?   The
"reality now" has quite widespread use of 6to4 (which is standardised) and
Teredo (which is not), although both have loosely similar architectural
implications.

> It would be much easier if more people did the work, instead of complained 
> about the work not having any result...

But as Christian and others point out, this is a potentially complex task,
most especially for enterprise.  One could see now where ISATAP provides
a unique solution for a certain sparse deployment in an enterprise, for
example.   Enterprise is the hardest of all four to nail down.

One could foresee that ISPs would need to deploy system(s) that act as
combined tunnel broker, Teredo relay and 6to4 relay as a "migration broker"
if we do ultimately end up going down those paths.   That's quite a price.
But I don't see the big archiecture discussion happening - I would have
thought this might be independent of the four (ISP/3GPP/unman/ent) scenarios?

Tim