Re: [v6ops] PMTUD forever, was: MTUs on the general Internet

[Jeroen Massar <[e-mail]>]

On 2014-12-05 12:45, Iljitsch van Beijnum wrote:
> On 05 Dec 2014, at 12:00, Philip Homburg <[e-mail]> wrote:
> 
>>> Only end hosts do fragmentation in IPv6.
> 
>> I was talking about tunnels. In general, tunnel endpoints are hosts.
> 
> Not sure if that's true. But if it is, there's no problem anyway, as the host will
>
> advertise an MSS that reflects the path MTU.

psst... MSS only applies to TCP.

We need a new term likely though for what you want to say:
 MPL = Max Payload Length

As that is what we are talking about, not the MTU.

> I.e., tunnel on host: IPv4 MTU = 1500, IPv6 MTU = 1480,
> IPv6 MSS = 1480 - 60 = 1420 and no too bigs are required.

Yep, you definitely mean MPL there.
(which might match MSS, but is not MSS)


The problem some folks in this discussion seem to want to avoid though
is that tunneled MTU deviates off 1500.

And they thus want to force tunnels to always have a MTU of 1500 and
thus force the lower layer to fragment packets instead.

While indeed then for that section you might not have any PMTU issues
anymore, you will create a whole bunch of overhead by fragmenting
packets (multiple packets, more possible packet loss, more traffic needs
to be sent etc). Next to the deployed base not support it anyway (see
other mail checking Linux code that is currently on github, not the
stuff that is even out on the Internet which can be ancient)...

Greets,
 Jeroen