Re: The IPv4 Internet MTU

[arno@natisbad.org (Arnaud Ebalard)]

Hi,

james woodyatt <jhw@apple.com> writes:

> [ snipped things i agree with ]
>
> Let's concentrate on the problems that belong to us.  That's what I say.

IMHO, the local issues can be solved ... locally, i.e. in home and
corporate networks.

Problems in the core are the real issue here: it's unbearable that
people from ISP or having some power on devices in the core implement
stupid filtering or deploy devices that simply do not work.

Let's correct that with AWARENESS.

What about simply:

1) pointing the finger at the people that do that: among the operator
   community, their reactivity (or lack of) will (slowly) create some
   feedback. For ISP, end users will simply know that things do not work
   smoothly with A and then select B. I don't know for peering.
2) referencing the brands that have devices that do that (i.e. bad
   implementations): the market will do the rest.

Let me be first: one of my ISP in France provides native IPv6 for their
ADSL service (they are the only one here, AFAIK). Basically, the
connectivity from the client's router to the LNS is provided either by
France Telecom or 9/Cegetel. When it's 9/Cegetel, you get an asymmetric
MTU: IPv6 packets larger than 1473 bytes sent by peers are just dropped
when they come back. Annoying to say the least ...

I have spent hours with people of my provider debugging that issue
(before switching to their SDSL service) and we get to the conclusion
that it was associated with the way IPv6 traffic encapsulation was done
by the devices in 9/Cegetel network. From what i get (_NO_ certainty on
that one), the issue is related to the L2TP implementation of HUAWEI
devices in 9/Cegetel network. I have put someone of my ISP in Bcc, he
might provide more specific feedback (or correct me).

Let's start referencing the faulty devices/networks/admins, this might
lead somewhere ... like having PMTUD working on the IPv6 internet. Note
that i don't think v6ops is the best place to do that.

a+

ps: in parallel, just like another client of my ISP (in Bcc), i
implemented TCP MSS clamping for IPv6 traffic for Linux but this is just
a workaround (IKE traffic obviously still get killed, for instance).