]> Git Repo - linux.git/commit
tcp: tso: allow CA_CWR state in tcp_tso_should_defer()
authorEric Dumazet <[email protected]>
Thu, 26 Feb 2015 22:10:20 +0000 (14:10 -0800)
committerDavid S. Miller <[email protected]>
Sat, 28 Feb 2015 20:10:39 +0000 (15:10 -0500)
commita0ea700e4095e26a8e291374757dd29519900f18
tree5dc19465848a991d8baf2ac3cac3079c92baef5f
parent50c8339e9299aaf09e48c8f94de39baec638c874
tcp: tso: allow CA_CWR state in tcp_tso_should_defer()

Another TCP issue is triggered by ECN.

Under pressure, receiver gets ECN marks, and send back ACK packets
with ECE TCP flag. Senders enter CA_CWR state.

In this state, tcp_tso_should_defer() is short cut :

if (icsk->icsk_ca_state != TCP_CA_Open)
    goto send_now;

This means that about all ACK packets we receive are triggering
a partial send, and because cwnd is kept small, we can only send
a small amount of data for each incoming ACK,
which in return generate more ACK packets.

Allowing CA_Open and CA_CWR states to enable TSO defer in
tcp_tso_should_defer() brings performance back :
TSO autodefer has more chance to defer under pressure.

This patch increases TSO and LRO/GRO efficiency back to normal levels,
and does not impact overall ECN behavior.

Signed-off-by: Eric Dumazet <[email protected]>
Signed-off-by: Yuchung Cheng <[email protected]>
Signed-off-by: Neal Cardwell <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
net/ipv4/tcp_output.c
This page took 0.054409 seconds and 4 git commands to generate.