MTU = 1280 everywhere? / QUIC

[Templin, Fred L]

Hi, the idea of setting a fixed 1280 MTU everywhere and for all time is silly;
the maximum MTU for IPv4 is 64KB, and the maximum MTU for IPv6 is 4GB.

One item of follow-up:

> Also, fragments are evil and there is no real reason to have any
> fragments at all.

IPv4 fragmentation works at slow speeds, but is dangerous at line rates.
IPv6 fragmentation works at line rates, but is a pain point that should be
avoided and/or tuned out when possible. Neither in and of themselves
are "evil", however.

Thanks - Fred
fred.l.templin at boeing.com

> -----Original Message-----
> From: ipv6-ops-bounces+fred.l.templin=boeing.com at lists.cluenet.de [mailto:ipv6-ops-
> bounces+fred.l.templin=boeing.com at lists.cluenet.de] On Behalf Of Jeroen Massar
> Sent: Tuesday, November 11, 2014 2:06 AM
> To: Vincent Bernat
> Cc: IPv6 Ops list
> Subject: Re: MTU = 1280 everywhere? / QUIC
> 
> On 2014-11-11 10:55, Vincent Bernat wrote:
> >  ❦ 11 novembre 2014 10:42 +0100, Jeroen Massar <jeroen at massar.ch> :
> >
> >> From:
> >> https://docs.google.com/document/d/1RNHkx_VvKWyWg6Lr8SZ-saqsQx7rFV-ev2jRFUoVD34/mobilebasic
> >> "UDP PACKET FRAGMENTATION" but IPv6 dos not fragment...
> >
> > IPv6 routers don't fragment but IPv6 hosts still do.
> 
> Correct. But that means if you are sending 1350 bytes on a 1280 link you
> are sending two packets, not one.
> 
> As they do cool stuff like FEC in QUIC, they assume lossy networks (good
> thing they think that way), but that also means that you will be sending
> more data (due to FEC) and also assume you are losing packets.
> 
> Hence, if your FEC protocol assumes that 1 packet is lost while actually
> only half the packet was, you got more loss than you are anticipating.
> 
> Knowing what the MTU is on the link, thus is a smart thing.
> 
> Hence, why PMTUD is important.
> 
> 
> Also, fragments are evil and there is no real reason to have any
> fragments at all.
> 
> Greets,
>  Jeroen