[dns-operations] L-Root Maintenance 2010-01-27 1800 UTC - 2000 UTC

[Rick Jones]

Robert Edmonds wrote:
> Duane Wessels wrote:
> 
>>When I mentally compare the "IP Protocols" and the "DNS Transport"
>>graphs, looks to me like its about 8 TCP packets per TCP DNS query...?
> 
> 
> minimum ought to be 7, right?
> 
> syn
>             syn+ack
> ack+query
>             ack+response
> fin+ack
>             fin+ack
> ack
> 

"It depends" - back in the days of HTTP 1.0 benchmarking, we were 
tweaking TCP stacks to get FIN's piggybacked on responses, of course 
HTTP 1.0 allowed that...since it let the server close after the first 
response.

In *theory* one could bundle the query syn and fin together, although 
the query would not be delivered until when, hmm, probably after the ACK 
of the syn|ack to avoid issues - although starting to service a query 
carried in the syn segment is no more spoofed than taking a query from a 
UDP datagram... so perhaps it is a don't care in the case of simple DNS 
stuff - in which case:

syn+query+fin
              syn+ack+response
fin+ack
              fin+ack
ack

you would need an eager connection notification on the server and a 
timer running to generate a syn|ack of the syn if generatig the response 
takes "too long"  it would also need a "connect with data" call at the 
client end.

but one would probably find all manner of state machine problems in 
stacks hither and yon across the Internet.

rick jones
probably missed something in there somewhere anyway...