Re: [BEHAVE] Fwd: IPv6 hosts sending <1280 byte packets

["Dan Wing" <[e-mail]>]

 

> -----Original Message-----
> From: Iljitsch van Beijnum [mailto:[e-mail]] 
> Sent: Tuesday, February 09, 2010 11:06 AM
> To: Dan Wing
> Cc: 'Templin, Fred L'; [e-mail]
> Subject: Re: [BEHAVE] Fwd: IPv6 hosts sending <1280 byte packets
> 
> On 9 feb 2010, at 19:50, Dan Wing wrote:
> 
> >> Only in the case where the IPv6 host supports an MTU larger 
> >> than 1500 but the translator doesn't, there is a possible 
> >> optimization by having the translator rewrite the MSS to the 
> >> maximum that the translator supports so the translator 
> >> wouldn't have to send too bigs and there is less risk of 
> >> PMTUD black holes. However, it's not clear to me that this 
> >> case is worth the trouble.
> 
> > There are long-standing desires to MTUs larger than 1500
> > with IPv6, under the theory that IPv6 Did Things Correctly.
> 
> You can send large packets with IPv4, too.

Yes, but experience on the IPv4 Internet shows that ICMP packet-too-big are
unreliable.  So many have given up on IPv4.  Closed environments, including
where tunnels are operating over equipment you own/control, are different.
But that is why I said 'desires' and put 
'Did Things Correctly' in quotes.  My goal, as chair, is to make sure
translators do not make deployment of large-MTU networks worse.  If they do,
it will complicate acceptance by IESG.

-d


> Having a bigger 
> than 1500 byte MTU causes much fewer problems than having a 
> smaller than 1500 byte MTU for a number of reasons which are 
> explained in http://tools.ietf.org/html/draft-van-beijnum-multi-mtu-02
> 
> However, if you have a 1500+ MTU on the IPv6 side (but the 
> translator has 1500), the MSS doesn't fix the PMTUD black 
> holes that are caused by the fact that that the IPv6 host has 
> a 1480 path MTU seen from the IPv4 side while it advertising 
> a larger one in the MSS.
> 
> If we want to avoid this particular black hole the solution 
> is to rewrite the MSS in the translator to the minimum of the 
> IPv6 host's MSS and the translator MTU on the IPv6 side + 
> sizeof(IPv4) - sizeof(IPv6).
> 
> I don't remember the discussion about MSS rewriting in the 
> softwires context, just that my conclusion was that that 
> issue doesn't apply here. In any event, MSS clamping has been 
> a staple of cheap NAT/PPPoE boxes and I don't believe it 
> causes any problems.=