netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: "John Heffner" <johnwheffner@gmail.com>
To: "Jerry Chu" <hkchu@google.com>
Cc: "David Miller" <davem@davemloft.net>,
	netdev@vger.kernel.org, rick.jones2@hp.com
Subject: Re: Socket buffer sizes with autotuning
Date: Mon, 28 Apr 2008 12:21:53 -0700	[thread overview]
Message-ID: <1e41a3230804281221s11442a82j55c363116988a5cb@mail.gmail.com> (raw)
In-Reply-To: <d1c2719f0804281130n7b0aaab8t54b2a585cff53a99@mail.gmail.com>

[-- Attachment #1: Type: text/plain, Size: 1036 bytes --]

On Mon, Apr 28, 2008 at 11:30 AM, Jerry Chu <hkchu@google.com> wrote:
>  Correct, but its counterpart doesn't exist in tcp_is_cwnd_limited(). So
>  cwnd will continue to grow when left < cwnd/sysctl_tcp_tso_win_divisor,
>  which can be very large when cwnd is large.
>
>  If I change tcp_tso_win_divisor to 0, cwnd max out at 695 rather than 1037,
>  exactly off by 1/3. I tried to add the same check to tcp_is_cwnd_limited():
>
>  diff -c /tmp/tcp.h.old tcp.h
>  *** /tmp/tcp.h.old      Mon Apr 28 11:00:44 2008
>  --- tcp.h       Mon Apr 28 10:54:10 2008
>  ***************
>  *** 828,833 ****
>  --- 828,835 ----
>                 return 0;
>
>         left = tp->snd_cwnd - in_flight;
>  +       if (left >= 65536)
>  +               return 0;
>         if (sysctl_tcp_tso_win_divisor)
>                 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
>         else
>
>  >
>
>  But it doesn't seem to help (cwnd still grows to 1037).

Cwnd is in packets, not bytes.

Try this series of patches, against net-next.

  -John

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-NET-Allow-send-limited-cwnd-to-grow-up-to-max_bur.patch --]
[-- Type: text/x-diff; name=0001-NET-Allow-send-limited-cwnd-to-grow-up-to-max_bur.patch, Size: 1231 bytes --]

From 57039cfc486ae4deac3b1fecaa238b5483c034df Mon Sep 17 00:00:00 2001
From: John Heffner <johnwheffner@gmail.com>
Date: Mon, 28 Apr 2008 12:17:49 -0700
Subject: [PATCH 1/2] [NET]: Allow send-limited cwnd to grow up to max_burst when gso disabled

This changes the logic in tcp_is_cwnd_limited() so that cwnd may grow
up to tcp_max_burst() even when sk_can_gso() is false, or when
sysctl_tcp_tso_win_divisor != 0.

Signed-off-by: John Heffner <johnwheffner@gmail.com>
---
 net/ipv4/tcp_cong.c |   11 ++++-------
 1 files changed, 4 insertions(+), 7 deletions(-)

diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
index 3a6be23..bfb1996 100644
--- a/net/ipv4/tcp_cong.c
+++ b/net/ipv4/tcp_cong.c
@@ -285,14 +285,11 @@ int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
 	if (in_flight >= tp->snd_cwnd)
 		return 1;
 
-	if (!sk_can_gso(sk))
-		return 0;
-
 	left = tp->snd_cwnd - in_flight;
-	if (sysctl_tcp_tso_win_divisor)
-		return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
-	else
-		return left <= tcp_max_burst(tp);
+	if (sk_can_gso(sk) &&
+	    left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd)
+		return 1;
+	return left <= tcp_max_burst(tp);
 }
 EXPORT_SYMBOL_GPL(tcp_is_cwnd_limited);
 
-- 
1.5.2.5


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #3: 0002-NET-Limit-cwnd-growth-when-deferring-for-GSO.patch --]
[-- Type: text/x-diff; name=0002-NET-Limit-cwnd-growth-when-deferring-for-GSO.patch, Size: 992 bytes --]

From fc1d9fe99748f3811a0a5b9d93463cf0e4788f56 Mon Sep 17 00:00:00 2001
From: John Heffner <johnwheffner@gmail.com>
Date: Mon, 28 Apr 2008 12:19:22 -0700
Subject: [PATCH 2/2] [NET]: Limit cwnd growth when deferring for GSO

This fixes inappropriately large cwnd growth on sender-limited flows
when GSO is enabled, limiting cwnd growth to 64k.

Signed-off-by: John Heffner <johnwheffner@gmail.com>
---
 net/ipv4/tcp_cong.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/net/ipv4/tcp_cong.c b/net/ipv4/tcp_cong.c
index bfb1996..6a25082 100644
--- a/net/ipv4/tcp_cong.c
+++ b/net/ipv4/tcp_cong.c
@@ -287,7 +287,8 @@ int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
 
 	left = tp->snd_cwnd - in_flight;
 	if (sk_can_gso(sk) &&
-	    left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd)
+	    left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd &&
+	    left * tp->mss_cache < sk->sk_gso_max_size)
 		return 1;
 	return left <= tcp_max_burst(tp);
 }
-- 
1.5.2.5


  reply	other threads:[~2008-04-28 19:22 UTC|newest]

Thread overview: 56+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-04-23 23:29 Socket buffer sizes with autotuning Jerry Chu
2008-04-24 16:32 ` John Heffner
2008-04-25  0:49   ` Jerry Chu
2008-04-25  6:46     ` David Miller
2008-04-25 21:29       ` Jerry Chu
2008-04-25 21:35         ` David Miller
2008-04-28 18:30       ` Jerry Chu
2008-04-28 19:21         ` John Heffner [this message]
2008-04-28 20:44           ` Jerry Chu
2008-04-28 23:22             ` [PATCH 1/2] [NET]: Allow send-limited cwnd to grow up to max_burst when gso disabled John Heffner
2008-04-28 23:22               ` [PATCH 2/2] [NET]: Limit cwnd growth when deferring for GSO John Heffner
     [not found]           ` <d1c2719f0804281338j3984cf2bga31def0c2c1192a1@mail.gmail.com>
2008-04-28 23:28             ` Socket buffer sizes with autotuning John Heffner
2008-04-28 23:35               ` David Miller
2008-04-29  2:20               ` Jerry Chu
2008-04-25  7:05 ` David Miller
2008-05-07  3:57   ` Jerry Chu
2008-05-07  4:27     ` David Miller
2008-05-07 18:36       ` Jerry Chu
2008-05-07 21:18         ` David Miller
2008-05-08  1:37           ` Jerry Chu
2008-05-08  1:43             ` David Miller
2008-05-08  3:33               ` Jerry Chu
2008-05-12 22:22                 ` Jerry Chu
2008-05-12 22:29                   ` David Miller
2008-05-12 22:31                     ` David Miller
2008-05-13  3:56                       ` Jerry Chu
2008-05-13  3:58                         ` David Miller
2008-05-13  4:00                           ` Jerry Chu
2008-05-13  4:02                             ` David Miller
2008-05-17  1:13                               ` Jerry Chu
2008-05-17  1:29                                 ` David Miller
2008-05-17  1:47                                   ` Jerry Chu
2008-05-12 22:58                     ` Jerry Chu
2008-05-12 23:01                       ` David Miller
2008-05-07  4:28     ` David Miller
2008-05-07 18:54       ` Jerry Chu
2008-05-07 21:20         ` David Miller
2008-05-08  0:16           ` Jerry Chu
     [not found] <d1c2719f0804241829s1bc3f41ejf7ebbff73ed96578@mail.gmail.com>
2008-04-25  7:06 ` Andi Kleen
2008-04-25  7:28   ` David Miller
2008-04-25  7:48     ` Andi Kleen
  -- strict thread matches above, loose matches on Subject: below --
2008-04-23  0:38 Rick Jones
2008-04-23  2:17 ` John Heffner
2008-04-23  3:59   ` David Miller
2008-04-23 16:32     ` Rick Jones
2008-04-23 16:58       ` John Heffner
2008-04-23 17:24         ` Rick Jones
2008-04-23 17:41           ` John Heffner
2008-04-23 17:46             ` Rick Jones
2008-04-24 22:21     ` Andi Kleen
2008-04-24 22:39       ` John Heffner
2008-04-25  1:28       ` David Miller
     [not found]       ` <65634d660804242234w66455bedve44801a98e3de9d9@mail.gmail.com>
2008-04-25  6:36         ` David Miller
2008-04-25  7:42           ` Tom Herbert
2008-04-25  7:46             ` David Miller
2008-04-28 17:51               ` Tom Herbert

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1e41a3230804281221s11442a82j55c363116988a5cb@mail.gmail.com \
    --to=johnwheffner@gmail.com \
    --cc=davem@davemloft.net \
    --cc=hkchu@google.com \
    --cc=netdev@vger.kernel.org \
    --cc=rick.jones2@hp.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).