Isn't one of the benefits of delayed ack that the acks get sent
back with the character echo (or other display update) for
interactive sessions?
Yes, in general, this is true and the intention of the behavior.
Maybe acks for short (<mss) frames should be delayed _less_ than
long (=mss) frames.
And this is precisely what my patch does, it delays, but for a very
short amount of time, yet always (in my experiments) enough time for
the application to send the echo back before the delayed ack timer
fires, so the ACK can piggyback the echo data packet.
I'm sure there is a small amount of tuning to do with my current
scheme, but for the most part I believe the heuristic to be sound,
again this is based upon my own experimentation and watching the wire
with tcpdump using my current TCP stack.
Later,
David S. Miller
davem@dm.cobaltmicro.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu