Slow start now increases cwnd by 1 if an ACK acknowledges some packets,
regardless the number of packets. Consequently slow start performance
is highly dependent on the degree of the stretch ACKs caused by
receiver or network ACK compression mechanisms (e.g., delayed-ACK,
GRO, etc). But slow start algorithm is to send twice the amount of
packets of packets left so it should process a stretch ACK of degree
N as if N ACKs of degree 1, then exits when cwnd exceeds ssthresh. A
follow up patch will use the remainder of the N (if greater than 1)
to adjust cwnd in the congestion avoidance phase.
In addition this patch retires the experimental limited slow start
(LSS) feature. LSS has multiple drawbacks but questionable benefit. The
fractional cwnd increase in LSS requires a loop in slow start even
though it's rarely used. Configuring such an increase step via a global
sysctl on different BDPS seems hard. Finally and most importantly the
slow start overshoot concern is now better covered by the Hybrid slow
start (hystart) enabled by default.
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
None of the existing TCP congestion controls use the rtt value pased
in the ca_ops->cong_avoid interface. Which is lucky because seq_rtt
could have been -1 when handling a duplicate ack.
Signed-off-by: Stephen Hemminger <shemminger@linux-foundation.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Reuse limited slow-start (RFC3742) included into tcp_cong instead
of having another implementation in High Speed TCP.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>
I think there is still a problem with the AIMD parameter update in
HighSpeed TCP code.
Line 125~138 of the code (net/ipv4/tcp_highspeed.c):
/* Update AIMD parameters */
if (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd) {
while (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd &&
ca->ai < HSTCP_AIMD_MAX - 1)
ca->ai++;
} else if (tp->snd_cwnd < hstcp_aimd_vals[ca->ai].cwnd) {
while (tp->snd_cwnd > hstcp_aimd_vals[ca->ai].cwnd &&
ca->ai > 0)
ca->ai--;
In fact, the second part (decreasing ca->ai) never decreases since the
while loop's inequality is in the reverse direction. This leads to
unfairness with multiple flows (once a flow happens to enjoy a higher
ca->ai, it keeps enjoying that even its cwnd decreases)
Here is a tentative fix (I also added a comment, trying to keep the
change clear):
Acked-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Implementation of RFC3742 limited slow start. Added as part
of the TCP highspeed congestion control module.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
When snd_cwnd is smaller than 38 and the connection is in
congestion avoidance phase (snd_cwnd > snd_ssthresh), the snd_cwnd
seems to stop growing.
The additive increase was confused because C array's are 0 based.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Xiaoliang (David) Wei wrote:
> Hi gurus,
>
> I am reading the code of tcp_highspeed.c in the kernel and have a
> question on the hstcp_cong_avoid function, specifically the following
> AI part (line 136~143 in net/ipv4/tcp_highspeed.c ):
>
> /* Do additive increase */
> if (tp->snd_cwnd < tp->snd_cwnd_clamp) {
> tp->snd_cwnd_cnt += ca->ai;
> if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
> tp->snd_cwnd++;
> tp->snd_cwnd_cnt -= tp->snd_cwnd;
> }
> }
>
> In this part, when (tp->snd_cwnd_cnt == tp->snd_cwnd),
> snd_cwnd_cnt will be -1... snd_cwnd_cnt is defined as u16, will this
> small chance of getting -1 becomes a problem?
> Shall we change it by reversing the order of the cwnd++ and cwnd_cnt -=
> cwnd?
Absolutely correct. Thanks.
Signed-off-by: John Heffner <jheffner@psc.edu>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is a compile error that crept in with the last patch of
TCP patches.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move all the code that does linear TCP slowstart to one
inline function to ease later patch to add ABC support.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
TCP peformance with TSO over networks with delay is awful.
On a 100Mbit link with 150ms delay, we get 4Mbits/sec with TSO and
50Mbits/sec without TSO.
The problem is with TSO, we intentionally do not keep the maximum
number of packets in flight to fill the window, we hold out to until
we can send a MSS chunk. But, we also don't update the congestion window
unless we have filled, as per RFC2861.
This patch replaces the check for the congestion window being full
with something smarter that accounts for TSO.
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
This changeset basically moves tcp_sk()->{ca_ops,ca_state,etc} to inet_csk(),
minimal renaming/moving done in this changeset to ease review.
Most of it is just changes of struct tcp_sock * to struct sock * parameters.
With this we move to a state closer to two interesting goals:
1. Generalisation of net/ipv4/tcp_diag.c, becoming inet_diag.c, being used
for any INET transport protocol that has struct inet_hashinfo and are
derived from struct inet_connection_sock. Keeps the userspace API, that will
just not display DCCP sockets, while newer versions of tools can support
DCCP.
2. INET generic transport pluggable Congestion Avoidance infrastructure, using
the current TCP CA infrastructure with DCCP.
Signed-off-by: Arnaldo Carvalho de Melo <acme@mandriva.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Sally Floyd's high speed TCP congestion control.
This is useful for comparison and research.
Signed-off-by: John Heffner <jheffner@psc.edu>
Signed-off-by: Stephen Hemminger <shemminger@osdl.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Yuchung Cheng