If we accept that 1280 is the only non-broken MTU for IPv6, I share
James' fear that this would spell doom for IPv6 on the broad Internet
and land us in a world where broad IPv6 adoption stalls out and IPv4
ith CGN dominates. The current reality is that many hosts do have
1500 byte MTUs (as Tore observed with many of the top Alexa sites, and
presumably also with the default configuration for many hosts out
there). We need to make sure that this current reality keeps working.
If content providers need to drop their IPv6 MTU/MSS down below that
of IPv4 (eg, to 1280), this has the potential to be enough of a
performance difference (even a few percent matters!) to cause many
content providers to hold off indefinitely on dual-stacking. If large
ISPs or enterprises need to find some way to get their end users down
to lower MTUs or need to wait for some OS updates to do some other
form of PMTU discovery, then IPv6 is DOA for them as well and they'll
have a good excuse to not roll out out.
As such, I agree with other comments that it's important
to re-emphasize that any environment that both blocks ICMPv6 PTB
and at the same time doesn't allow 1500 byte packets through is broken
and needs to get fixed. This also means that anyone deploying
or implementing a tunnel (or VPN) needs to carefully consider
how they'll deal with this.
For fragments themselves, would it make more sense to constrain
their usage rather than deprecate them entirely? For example,
express that certain headers may not appear in fragments,
and also perhaps express that overlapping fragments or overly
spase fragments are likely to be dropped?
In looking at a global end-user facing HTTP (TCP) environment,
I'm seeing around 0.0005% of the packets arrive as fragments
which sounds similar to some of the other numbers on this thread.
Erik
On Tue, Jun 25, 2013 at 7:48 PM, Brian E Carpenter
<brian.e.carpen...@gmail.com> wrote:
> On 26/06/2013 11:08, james woodyatt wrote:
>> On Jun 25, 2013, at 13:13 , Randy Bush <ra...@psg.com> wrote:
>>> ron has made one suggestion. though brutal, its simplicity and its
>>> recognition of reality appeal to me.
>>
>> Why not go large and deprecate all of RFC 2460?
>>
>> I'm not entirely joking. If the practical global MTU for IPv6 is 1280, ...
>
> We went through a protracted phase where the practical global MTU
> for IPv4 was around 512, unless you wanted to reconfigure your laptop
> stack experimentally each time you checked into a hotel and figured
> out how to connect to the phone line. I even recall having to set the
> MTU to 256 in one hotel. We may well have to accept a protracted phase
> of 1280 for IPv6. Eventually it will resolve itself, as it did for IPv4.
> I think that's orthogonal to the question of deprecating L3 fragmentation.
>
> Brian
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> ipv6@ietf.org
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------
--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
