Re: The IPv4 Internet MTU

[Magnus Westerlund <magnus.westerlund@ericsson.com>]

Iljitsch van Beijnum skrev:
> On 10-okt-2007, at 22:51, Brian E Carpenter wrote:
> 
>> On 2007-10-11 00:56, Magnus Westerlund wrote:
>>> Especially for UDP applications that rarely has
>>> built in any support for handling MTU discovery if the ICMP doesn't
>>> work.
> 
> Are we talking about IPv4 here? In that case, path MTU discovery is
> pretty much impossible because the applications would have to receive
> the ICMP messages and adjust their behavior = every UDP app needs to
> implement PMTUD itself.

Yes IPv4, but I don't know if things are better in the IPv6 world. What
you write about requiring each app to do it itself is basically the
state of things today as I understand them. However, you should even
forget to use ICMP for it. It can be used as a hint for your process.
But the only reliable way of determining MTU is to probe how big packets
the network will deliver with DF bit set. And this obviously doesn't
work unless you have a feedback and can ACK which packets that was
delivered so that one can know if a particular size successfully was
delivered.

> 
> With IPv6 the IP layer takes care of all of this but obviously if there
> are no ICMP messages it won't work. However...
> 
>>> At least for TCP we at least have a solution that can be gradually
>>> deployed in RFC 4821 "Packetization Layer Path MTU Discovery" that
>>> doesn't rely on ICMP.
> 
> I don't see any reason why the IPv6 IP layer wouldn't be able to do RFC
> 4821.
> 
>>> However, I think we need to strive towards getting applications in line
>>> and support MTU discovery.
> 
> No way. Application developers don't run networks, they tend to get this
> stuff wrong and then you'll forever supporting applications with broken
> network behavior that won't be upgraded for a decade.

Well, then we stop using UDP. Protocols built on top of UDP needs to
take care of this as ICMP clearly does not work. There is also the whole
thing that most applications need to deliver the data that suits the MTU
that is present. For RTP/UDP application this adaptation needs to happen
in the media codecs that deliver data to the lower layers to enable the
utilization of different MTUs.

Any protocol/application implementation capable of utilizing larger MTUs
needs to have support for utilizing the MTU that actually is present.


>>> It is silly for high-bit rate applications to
>>> send all these small packet instead of at least descent sized packets
>>> when possible.
> 
>> So, is there any scope for a generic UDP echo process that would allow
>> PMTUD without ICMP messages?
> 
> That's what RFC 4821 does by trying different packet sizes and seeing which gets through.
> 
> For IPv4, an obvious method would be to send packets with DF=0 and hear from the other side how big the fragments are but that's not possible with IPv4.
> 

Yes, that would probably work. The question is how we get a feedback
channel that works.

Cheers

Magnus Westerlund

IETF Transport Area Director & TSVWG Chair
----------------------------------------------------------------------
Multimedia Technologies, Ericsson Research EAB/TVM/M
----------------------------------------------------------------------
Ericsson AB                | Phone +46 8 4048287
Torshamsgatan 23           | Fax   +46 8 7575550
S-164 80 Stockholm, Sweden | mailto: magnus.westerlund@ericsson.com
----------------------------------------------------------------------