Mark Smith;
Radia Perlman, in her book "Interconnections", 2nd edition,
suggests a few alternative methods of performing PMTUD, including
one which wouldn't require feedback from the network, starting at
pg 185.
While I won't bother to read a paper material not available here,
I understand it is a very bad idea for a transport or above layer
to interpret a packet loss as a unreliable hint of congestion and
MTU mismatch, especially because initially and upon path changes,
both path congestion and path MTU properties will change.
It is fine that fragmentation of IPv4 is detected and reacted
against by an upper layer.
It is also fine to just send 1280B or smaller IPv6 packets.
I'm not sure I understand you. Are you saying the idea of PMTUD
is broken, or the way it currently works ?
The idea is broken. Note that recent super computers are, as
expected long ago, massively parallel ones to be able to
process a lot of interrupts without much performance loss.
Masataka Ohta
_______________________________________________
Ietf mailing list
Ietf(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/ietf