[dns-operations] PMTUD of .org servers

[Rick Jones]

> at which point EDNS will suffer badly, and TCP fallback will be used.  so
> like i said, solaris just can't be a good EDNS server.  interesting.

I'm guessing then that HP-UX 11i will be in the same boat.

Since I'm more than willing to demonstrate my cluelessness :)  why is it 
felt that the occasional drop from PMTU will cause such a degree of 
suffering?  Doesn't it depend on the frequency of queries relative to 
the PMTU entry timeouts? (ip_ire_pathmtu_interval in HP-UX ndd-speak).

And is wiretapping the ICMP really going to help much in named 
(udp_pass_up_icmp in HP-UX nddspeak)?  IIRC on those hosts where DF is 
set on UDP, the arrival of the ICMP message is going to cause a PMTU 
route to be created and there will be suitable IP fragmentation on the 
next send(s) to that destination.  Unless named is going to keep a copy 
of a reply on the off chance an ICMP is going to arrive it doesn't 
really benefit from getting the ICMP message(s) does it?

I suppose when named is sending a request there might be value, but is 
that going to be worth the overheads?

rick jones