Yes. On a 30mS round trip time the fact Nagle kicks in between headers
and body just took you out of the web server benchmarks completely. You
are now 30mS slower clearing a service, your number of threads has
gone up, your live L1/L2 cache has more in it and your context switch
rate has gone up.
Once you factor in transatlantic paths - 100mS or more Nagle is a
huge deal. Even the TCP slow start behaviour has become a big enough
deal the IETF working group went through months of arguing, simulation
and proving to accept a bigger initial congestion window was valid
Alan
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu