[PATCH] Added NoDelay config option and nodelay subsystem option

[Rick Jones]

Kevin Steves wrote:
> :Hmm, an interesting comment to leave dangling out there, but probably
> :best for tcp-impl :)
> 
> well, most issues seem to stem from poor intereaction with fixed-timer
> delayed ACKs.  are any stack doing something more intelligent?

Only if they can be clairvoyant. One might initially say that any
sub-MSS send should get an immediate ACK, but then you start sending all
these otherwise bogus ACKs for otherwise correctly behaving apps - for
example, the request/response of a single-byte (sub-mss) netperf TCP_RR
test (think really fast typist :) would go from two TCP segments per
transaction to four. That would basically double the CPU time spent in
the networking stack.

The stack would need to know whether or not the recieving app was going
to send data the other way any time soon and whether or not the other
side had more data to send.

The best one can do from what I can think of as I type is to use a
shorter standalone ACK interval and work to make sure that apps which
interact poorly with Nagle are either writen to behave properly or if
they are already behaving properly to (as much as I hate to say) disable
Nagle.

rick jones
-- 
Wisdom Teeth are impacted, people are affected by the effects of events.
these opinions are mine, all mine; HP might not want them anyway... :)
feel free to post, OR email to raj in cup.hp.com  but NOT BOTH...