Re: draft-wbeebee-ipv6-cpe-router-04 comments

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 20 jul 2009, at 18:23, Mikael Abrahamsson wrote:

> > Why do you guess this?

> Because the people who mess up usually have larger packet sizes than  
> 1350

But why 1350 and not 1349 or 1351?

> because the brokenness comes from tunnels.

Brokenness comes from people filtering ICMP but still having PMTUD  
enabled.

> much more likely to get it right than the guys who try to run 1500  
> and get it wrong.

You confuse "works" with "right". That's exactly how the problem you  
try to solve started. While I can imagine operators displaying this  
level of pragmatism, we have to do better in the IETF.

> So the question is, do we try to work around the brokenness or do we  
> just do the right thing ourselves and call everybody else "broken"  
> and sit there with a less working IPv6 than we could have.

Going along with people who break the specifications is a road that  
leads nowehere fast.

The problem with setting an arbitrary packet size is not just that  
it's a race to the bottom (of course now someone runs a tunnel over a  
link with a 1350 MTU so you need to use a 1330 MTU etc etc) but also  
that there is no exit strategy. Even if you upgrade everything to use  
a link layer that supports bigger packets, you're still sending 1350- 
byte packets.

Now at least that last problem is solvable by not hardcoding "1350"  
but "link layer MTU - IPv4 (+IPsec+...) header".

> So I think the question is, is this document about doing it right or  
> getting it to work to the largest extent as possible?

See the E in IETF.