diff mbox series

[net-next,v3] tcp: Set pingpong threshold via sysctl

Message ID 1697056244-21888-1-git-send-email-haiyangz@microsoft.com
State New
Headers
Series [net-next,v3] tcp: Set pingpong threshold via sysctl |

Commit Message

Haiyang Zhang Oct. 11, 2023, 8:30 p.m. UTC 
  TCP pingpong threshold is 1 by default. But some applications, like SQL DB
may prefer a higher pingpong threshold to activate delayed acks in quick
ack mode for better performance.

The pingpong threshold and related code were changed to 3 in the year
2019 in:
  commit 4a41f453bedf ("tcp: change pingpong threshold to 3")
And reverted to 1 in the year 2022 in:
  commit 4d8f24eeedc5 ("Revert "tcp: change pingpong threshold to 3"")

There is no single value that fits all applications.
Add net.ipv4.tcp_pingpong_thresh sysctl tunable, so it can be tuned for
optimal performance based on the application needs.

Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
---
v3: Updated doc as suggested by Neal Cardwell.
    Updated variable location in struct netns_ipv4 as suggested by Kuniyuki
    Iwashima.

v2: Make it per-namesapce setting, and other updates suggested by Neal Cardwell,
and Kuniyuki Iwashima.
---
 Documentation/networking/ip-sysctl.rst | 13 +++++++++++++
 include/net/inet_connection_sock.h     | 16 ++++++++++++----
 include/net/netns/ipv4.h               |  2 ++
 net/ipv4/sysctl_net_ipv4.c             |  8 ++++++++
 net/ipv4/tcp_ipv4.c                    |  2 ++
 net/ipv4/tcp_output.c                  |  4 ++--
 6 files changed, 39 insertions(+), 6 deletions(-)

Comments

Simon Horman Oct. 16, 2023, 9:10 a.m. UTC | #1 
On Wed, Oct 11, 2023 at 01:30:44PM -0700, Haiyang Zhang wrote:
> TCP pingpong threshold is 1 by default. But some applications, like SQL DB
> may prefer a higher pingpong threshold to activate delayed acks in quick
> ack mode for better performance.
> 
> The pingpong threshold and related code were changed to 3 in the year
> 2019 in:
>   commit 4a41f453bedf ("tcp: change pingpong threshold to 3")
> And reverted to 1 in the year 2022 in:
>   commit 4d8f24eeedc5 ("Revert "tcp: change pingpong threshold to 3"")
> 
> There is no single value that fits all applications.
> Add net.ipv4.tcp_pingpong_thresh sysctl tunable, so it can be tuned for
> optimal performance based on the application needs.
> 
> Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
> ---
> v3: Updated doc as suggested by Neal Cardwell.
>     Updated variable location in struct netns_ipv4 as suggested by Kuniyuki
>     Iwashima.
> 
> v2: Make it per-namesapce setting, and other updates suggested by Neal Cardwell,
> and Kuniyuki Iwashima.

Thanks,

this looks clean to me. It seems to address the review of v2.
And keeps the knob as syctl as discussed in v2.

Reviewed-by: Simon Horman <horms@kernel.org>
Eric Dumazet Oct. 16, 2023, 11:40 a.m. UTC | #2 
On Wed, Oct 11, 2023 at 10:31 PM Haiyang Zhang <haiyangz@microsoft.com> wrote:
>
> TCP pingpong threshold is 1 by default. But some applications, like SQL DB
> may prefer a higher pingpong threshold to activate delayed acks in quick
> ack mode for better performance.
>

...

>
> diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> index f207712eece1..7d0fe76d56ef 100644
> --- a/net/ipv4/tcp_output.c
> +++ b/net/ipv4/tcp_output.c
> @@ -170,10 +170,10 @@ static void tcp_event_data_sent(struct tcp_sock *tp,
>         tp->lsndtime = now;
>
>         /* If it is a reply for ato after last received
> -        * packet, enter pingpong mode.
> +        * packet, increase pingpong count.
>          */
>         if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
> -               inet_csk_enter_pingpong_mode(sk);
> +               inet_csk_inc_pingpong_cnt(sk);
>  }
>
>  /* Account for an ACK we sent. */

OK, but I do not think we solved the fundamental problem of using
jiffies for this heuristic,
especially for HZ=100 or HZ=250 builds.

Reviewed-by: Eric Dumazet <edumazet@google.com>
Neal Cardwell Oct. 16, 2023, 4:17 p.m. UTC | #3 
On Wed, Oct 11, 2023 at 4:31 PM Haiyang Zhang <haiyangz@microsoft.com> wrote:
>
> TCP pingpong threshold is 1 by default. But some applications, like SQL DB
> may prefer a higher pingpong threshold to activate delayed acks in quick
> ack mode for better performance.
>
> The pingpong threshold and related code were changed to 3 in the year
> 2019 in:
>   commit 4a41f453bedf ("tcp: change pingpong threshold to 3")
> And reverted to 1 in the year 2022 in:
>   commit 4d8f24eeedc5 ("Revert "tcp: change pingpong threshold to 3"")
>
> There is no single value that fits all applications.
> Add net.ipv4.tcp_pingpong_thresh sysctl tunable, so it can be tuned for
> optimal performance based on the application needs.
>
> Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>
> ---
> v3: Updated doc as suggested by Neal Cardwell.
>     Updated variable location in struct netns_ipv4 as suggested by Kuniyuki
>     Iwashima.
>
> v2: Make it per-namesapce setting, and other updates suggested by Neal Cardwell,
> and Kuniyuki Iwashima.
> ---
>  Documentation/networking/ip-sysctl.rst | 13 +++++++++++++
>  include/net/inet_connection_sock.h     | 16 ++++++++++++----
>  include/net/netns/ipv4.h               |  2 ++
>  net/ipv4/sysctl_net_ipv4.c             |  8 ++++++++
>  net/ipv4/tcp_ipv4.c                    |  2 ++
>  net/ipv4/tcp_output.c                  |  4 ++--
>  6 files changed, 39 insertions(+), 6 deletions(-)

Acked-by: Neal Cardwell <ncardwell@google.com>

Thanks!

neal
Kuniyuki Iwashima Oct. 16, 2023, 4:49 p.m. UTC | #4 
From: Haiyang Zhang <haiyangz@microsoft.com>
Date: Wed, 11 Oct 2023 13:30:44 -0700
> TCP pingpong threshold is 1 by default. But some applications, like SQL DB
> may prefer a higher pingpong threshold to activate delayed acks in quick
> ack mode for better performance.
> 
> The pingpong threshold and related code were changed to 3 in the year
> 2019 in:
>   commit 4a41f453bedf ("tcp: change pingpong threshold to 3")
> And reverted to 1 in the year 2022 in:
>   commit 4d8f24eeedc5 ("Revert "tcp: change pingpong threshold to 3"")
> 
> There is no single value that fits all applications.
> Add net.ipv4.tcp_pingpong_thresh sysctl tunable, so it can be tuned for
> optimal performance based on the application needs.
> 
> Signed-off-by: Haiyang Zhang <haiyangz@microsoft.com>

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>

Thanks!
patchwork-bot+netdevbpf@kernel.org Oct. 16, 2023, 10:30 p.m. UTC | #5 
Hello:

This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Wed, 11 Oct 2023 13:30:44 -0700 you wrote:
> TCP pingpong threshold is 1 by default. But some applications, like SQL DB
> may prefer a higher pingpong threshold to activate delayed acks in quick
> ack mode for better performance.
> 
> The pingpong threshold and related code were changed to 3 in the year
> 2019 in:
>   commit 4a41f453bedf ("tcp: change pingpong threshold to 3")
> And reverted to 1 in the year 2022 in:
>   commit 4d8f24eeedc5 ("Revert "tcp: change pingpong threshold to 3"")
> 
> [...]

Here is the summary with links:
  - [net-next,v3] tcp: Set pingpong threshold via sysctl
    https://git.kernel.org/netdev/net-next/c/562b1fdf061b

You are awesome, thank you!
diff mbox series

Patch

diff --git a/Documentation/networking/ip-sysctl.rst b/Documentation/networking/ip-sysctl.rst
index f7dfde3b09a9..e7ec9026e5db 100644
--- a/Documentation/networking/ip-sysctl.rst
+++ b/Documentation/networking/ip-sysctl.rst
@@ -1183,6 +1183,19 @@  tcp_plb_cong_thresh - INTEGER
 
 	Default: 128
 
+tcp_pingpong_thresh - INTEGER
+	The number of estimated data replies sent for estimated incoming data
+	requests that must happen before TCP considers that a connection is a
+	"ping-pong" (request-response) connection for which delayed
+	acknowledgments can provide benefits.
+
+	This threshold is 1 by default, but some applications may need a higher
+	threshold for optimal performance.
+
+	Possible Values: 1 - 255
+
+	Default: 1
+
 UDP variables
 =============
 
diff --git a/include/net/inet_connection_sock.h b/include/net/inet_connection_sock.h
index d6d9d1c1985a..086d1193c9ef 100644
--- a/include/net/inet_connection_sock.h
+++ b/include/net/inet_connection_sock.h
@@ -328,11 +328,10 @@  void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
 
 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu);
 
-#define TCP_PINGPONG_THRESH	1
-
 static inline void inet_csk_enter_pingpong_mode(struct sock *sk)
 {
-	inet_csk(sk)->icsk_ack.pingpong = TCP_PINGPONG_THRESH;
+	inet_csk(sk)->icsk_ack.pingpong =
+		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
 }
 
 static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
@@ -342,7 +341,16 @@  static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
 
 static inline bool inet_csk_in_pingpong_mode(struct sock *sk)
 {
-	return inet_csk(sk)->icsk_ack.pingpong >= TCP_PINGPONG_THRESH;
+	return inet_csk(sk)->icsk_ack.pingpong >=
+	       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
+}
+
+static inline void inet_csk_inc_pingpong_cnt(struct sock *sk)
+{
+	struct inet_connection_sock *icsk = inet_csk(sk);
+
+	if (icsk->icsk_ack.pingpong < U8_MAX)
+		icsk->icsk_ack.pingpong++;
 }
 
 static inline bool inet_csk_has_ulp(const struct sock *sk)
diff --git a/include/net/netns/ipv4.h b/include/net/netns/ipv4.h
index d96d05b08819..73f43f699199 100644
--- a/include/net/netns/ipv4.h
+++ b/include/net/netns/ipv4.h
@@ -133,6 +133,8 @@  struct netns_ipv4 {
 	u8 sysctl_tcp_migrate_req;
 	u8 sysctl_tcp_comp_sack_nr;
 	u8 sysctl_tcp_backlog_ack_defer;
+	u8 sysctl_tcp_pingpong_thresh;
+
 	int sysctl_tcp_reordering;
 	u8 sysctl_tcp_retries1;
 	u8 sysctl_tcp_retries2;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index e7f024d93572..f63a545a7374 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -1498,6 +1498,14 @@  static struct ctl_table ipv4_net_table[] = {
 		.extra1		= SYSCTL_ZERO,
 		.extra2		= SYSCTL_ONE,
 	},
+	{
+		.procname	= "tcp_pingpong_thresh",
+		.data		= &init_net.ipv4.sysctl_tcp_pingpong_thresh,
+		.maxlen		= sizeof(u8),
+		.mode		= 0644,
+		.proc_handler	= proc_dou8vec_minmax,
+		.extra1		= SYSCTL_ONE,
+	},
 	{ }
 };
 
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index a441740616d7..f603ad9307af 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -3288,6 +3288,8 @@  static int __net_init tcp_sk_init(struct net *net)
 	net->ipv4.sysctl_tcp_syn_linear_timeouts = 4;
 	net->ipv4.sysctl_tcp_shrink_window = 0;
 
+	net->ipv4.sysctl_tcp_pingpong_thresh = 1;
+
 	return 0;
 }
 
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index f207712eece1..7d0fe76d56ef 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -170,10 +170,10 @@  static void tcp_event_data_sent(struct tcp_sock *tp,
 	tp->lsndtime = now;
 
 	/* If it is a reply for ato after last received
-	 * packet, enter pingpong mode.
+	 * packet, increase pingpong count.
 	 */
 	if ((u32)(now - icsk->icsk_ack.lrcvtime) < icsk->icsk_ack.ato)
-		inet_csk_enter_pingpong_mode(sk);
+		inet_csk_inc_pingpong_cnt(sk);
 }
 
 /* Account for an ACK we sent. */