Permalink
Browse files

tcp_cubic: better follow cubic curve after idle period

Jana Iyengar found an interesting issue on CUBIC :

The epoch is only updated/reset initially and when experiencing losses.
The delta "t" of now - epoch_start can be arbitrary large after app idle
as well as the bic_target. Consequentially the slope (inverse of
ca->cnt) would be really large, and eventually ca->cnt would be
lower-bounded in the end to 2 to have delayed-ACK slow-start behavior.

This particularly shows up when slow_start_after_idle is disabled
as a dangerous cwnd inflation (1.5 x RTT) after few seconds of idle
time.

Jana initial fix was to reset epoch_start if app limited,
but Neal pointed out it would ask the CUBIC algorithm to recalculate the
curve so that we again start growing steeply upward from where cwnd is
now (as CUBIC does just after a loss). Ideally we'd want the cwnd growth
curve to be the same shape, just shifted later in time by the amount of
the idle period.

Reported-by: Jana Iyengar <jri@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Sangtae Ha <sangtae.ha@gmail.com>
Cc: Lawrence Brakmo <lawrence@brakmo.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
  • Loading branch information...
1 parent 05c5a46 commit 30927520dbae297182990bb21d08762bcc35ce1d Eric Dumazet committed with davem330 Sep 10, 2015
Showing with 16 additions and 0 deletions.
  1. +16 −0 net/ipv4/tcp_cubic.c
View
@@ -151,6 +151,21 @@ static void bictcp_init(struct sock *sk)
tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
}
+static void bictcp_cwnd_event(struct sock *sk, enum tcp_ca_event event)
+{
+ if (event == CA_EVENT_TX_START) {
+ s32 delta = tcp_time_stamp - tcp_sk(sk)->lsndtime;
+ struct bictcp *ca = inet_csk_ca(sk);
+
+ /* We were application limited (idle) for a while.
+ * Shift epoch_start to keep cwnd growth to cubic curve.
+ */
+ if (ca->epoch_start && delta > 0)
+ ca->epoch_start += delta;
+ return;
+ }
+}
+
/* calculate the cubic root of x using a table lookup followed by one
* Newton-Raphson iteration.
* Avg err ~= 0.195%
@@ -450,6 +465,7 @@ static struct tcp_congestion_ops cubictcp __read_mostly = {
.cong_avoid = bictcp_cong_avoid,
.set_state = bictcp_state,
.undo_cwnd = bictcp_undo_cwnd,
+ .cwnd_event = bictcp_cwnd_event,
.pkts_acked = bictcp_acked,
.owner = THIS_MODULE,
.name = "cubic",

41 comments on commit 3092752

@zhouyongtao

不明觉厉的样子

@shaoyongyang

what mean ?

@shaoyongyang

what mean ?

@shaoyongyang

what mean ?

@shaoyongyang

what mean?

@shaoyongyang

what mean?

@shaoyongyang

what mean?

@shaoyongyang

what mean?

@shaoyongyang

mean ?

@shaoyongyang

what man?

@vingc
vingc commented on 3092752 Sep 28, 2015

膜拜,感谢贡献者

@douglarek

It's Cool, thanks

@anythink-wx

顶~

@openproject

So good!

@Honghe
Honghe commented on 3092752 Sep 28, 2015

对 cubic algorithm 不知,因此看不懂此重要修复,可惜了

@xormplus

good job

@teddysun

Signed-off-by: Yuchung Cheng ycheng@google.com

I guess, this is a Chinese name.
Good job.

@lebiru
lebiru commented on 3092752 Sep 28, 2015

awesome 👍

@swaechter

Is the return inside the if statement really needed?

@liuchangdong

good job

@porscheyin

Yuchung Cheng is from Taiwan

@name327

我赵日天并不服

@kenmux
kenmux commented on 3092752 Sep 29, 2015

Why mention about this man specially? Why?
They're all from Google. Good jobs, guys!

@dhdone
dhdone commented on 3092752 Sep 29, 2015

感谢修复者

@zhenhua

How much performance improvement could we have by applying this fix? Could anyone give us a hint?

@marctmiller
@hephaex
@guhb
guhb commented on 3092752 Sep 29, 2015

看不懂~先顶一个

@kbengine

wow!

@rafavg77

Good Job!

@LINGQ1991

不明觉厉!!!

@gbjtv
gbjtv commented on 3092752 Oct 6, 2015

Will this make it into other lts kernels?

@theharveyz

awesome!

@dale76uk

Hi,
I'm running linux mint 17.1, how do I apply this patch?
Thanks in advance

@marctmiller
@nkeck720
@dale76uk

marc, there is no need to be so rude! I'm not that naive.

@marctmiller
@arjun024
Contributor

👏

@chandlerding

Well , we just need wait for redhat to backport this patch and roll up the new kernel updates...

@notaweelos

Thanks.

Please sign in to comment.