tcp: adjust tail loss probe timeout
authorYuchung Cheng <ycheng@google.com>
Wed, 19 Jul 2017 22:41:26 +0000 (15:41 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 19 Jul 2017 23:14:10 +0000 (16:14 -0700)
commitbb4d991a28cc86a2dfbeefeff32911ca9f779c18
tree6fd415a7d0b54a9e639067375013fe8f2c3a48b7
parentc4b2bf6b4a35348fe6d1eb06928eb68d7b9d99a9
tcp: adjust tail loss probe timeout

This patch adjusts the timeout formula to schedule the TCP loss probe
(TLP). The previous formula uses 2*SRTT or 1.5*RTT + DelayACKMax if
only one packet is in flight. It keeps a lower bound of 10 msec which
is too large for short RTT connections (e.g. within a data-center).
The new formula = 2*RTT + (inflight == 1 ? 200ms : 2ticks) which
performs better for short and fast connections.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/tcp.h
net/ipv4/tcp_output.c
net/ipv4/tcp_recovery.c