[Bloat] Fwd: Initial Spreading patch

[Bloat] Fwd: Initial Spreading patch

[renaud sallantin]

[-- Attachment #1: Type: text/plain, Size: 1901 bytes --]

Dear Dave,

A quick answer from a small paradise where I am ending my hollydays
(Internet is sadly not as efficient everywhere, and I am lucky to get your
email...) lets speak more about it next week !!!

Attached to this email, you can find our patch tested with Linux 3.13.8.
As you can see, we used a slightly modified version of the FQ scheduler
developed by Eric Dumazet.

We introduced two sysctl parameters to ease your tests:
- tcp_initial_spreading_rate_min (integer):
It corresponds to the minimal rate value computed by Initial Spreading.
Considering draft-sallantin-tcpm-initial-spreading-00,
                        tcp_initial_spreading_rate_min= MTU / T_Spreading

- tcp_initial_spreading_rate_debug (boolean):
When non-zero, a kernel log  message will print the rate used by Initial
Spreading each time a TCP connection is opened.

Regards,
Renaud Sallantin
Dear all,

Following our last email,
some of you asked for our Initial Spreading implementation.

Attached to this email, you can find our patch tested with Linux 3.13.8.
As you can see, we used a slightly modified version of the FQ scheduler
developed by Eric Dumazet.

We introduced two sysctl parameters to ease your tests:
- tcp_initial_spreading_rate_min (integer):
It corresponds to the minimal rate value computed by Initial Spreading.
Considering draft-sallantin-tcpm-initial-spreading-00,
                        tcp_initial_spreading_rate_min= MTU / T_Spreading

- tcp_initial_spreading_rate_debug (boolean):
When non-zero, a kernel log  message will print the rate used by Initial
Spreading each time a TCP connection is initialized


We hope that you will be able to try Initial Spreading and that you will
confirm the good results we obtained.

Your comments on the patch but also on the draft are more than welcome,
https://tools.ietf.org/html/draft-sallantin-tcpm-initial-spreading-00

Regards,
Renaud Sallantin

[-- Attachment #2: Type: text/html, Size: 2568 bytes --]

[Bloat] Initial Spreading patch

[renaud sallantin]

[-- Attachment #1.1: Type: text/plain, Size: 1457 bytes --]

Sorry,
This time it should work.

It looks like we have the same kind of vacations...
---------- Message transféré ----------
De : "renaud sallantin" <renaud.sallantin@gmail.com>
Date : 26 janv. 2015 10:13
Objet : Initial Spreading patch
À : "tcpm@ietf.org" <tcpm@ietf.org>, "cedric baudoin" <
cedric.baudoin@thalesaleniaspace.com>
Cc :

Dear all,
>
> Following our last email,
> some of you asked for our Initial Spreading implementation.
>
> Attached to this email, you can find our patch tested with Linux 3.13.8.
> As you can see, we used a slightly modified version of the FQ scheduler
> developed by Eric Dumazet.
>
> We introduced two sysctl parameters to ease your tests:
> - tcp_initial_spreading_rate_min (integer):
> It corresponds to the minimal rate value computed by Initial Spreading.
> Considering draft-sallantin-tcpm-initial-spreading-00,
>                         tcp_initial_spreading_rate_min= MTU / T_Spreading
>
> - tcp_initial_spreading_rate_debug (boolean):
> When non-zero, a kernel log  message will print the rate used by Initial
> Spreading each time a TCP connection is initialized
>
>
> We hope that you will be able to try Initial Spreading and that you will
> confirm the good results we obtained.
>
> Your comments on the patch but also on the draft are more than welcome,
> https://tools.ietf.org/html/draft-sallantin-tcpm-initial-spreading-00
>
> Regards,
> Renaud Sallantin
>

[-- Attachment #1.2: Type: text/html, Size: 2200 bytes --]

[-- Attachment #2: 0001-implementing-initial-spreading-using-fq-sched.patch --]
[-- Type: text/x-patch, Size: 5967 bytes --]

From 651f8c07ba4fb8e1ccdfaa9cc2eb94b51c902544 Mon Sep 17 00:00:00 2001
From: Elie Bouttier <elie.bouttier@enseeiht.fr>
Date: Thu, 11 Dec 2014 19:02:17 +0100
Subject: [PATCH] implementing initial spreading using fq sched

---
 Documentation/networking/ip-sysctl.txt | 13 +++++++++++++
 include/net/tcp.h                      |  2 ++
 include/uapi/linux/sysctl.h            |  2 ++
 net/ipv4/sysctl_net_ipv4.c             | 15 +++++++++++++++
 net/ipv4/tcp_input.c                   | 33 ++++++++++++++++++++++++++++++++-
 net/sched/sch_fq.c                     |  5 +++--
 6 files changed, 67 insertions(+), 3 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index 8a984e9..cbe1c39 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -588,6 +588,19 @@ tcp_challenge_ack_limit - INTEGER
 	in RFC 5961 (Improving TCP's Robustness to Blind In-Window Attacks)
 	Default: 100
 
+tcp_initial_spreading_min_rate - INTEGER
+    Minimal rate value computed by initial spreading algorithm.
+    Disabled if set to zero.
+    See
+    https://tools.ietf.org/html/draft-sallantin-tcpm-initial-spreading-00.
+    We have the relation rate_min = MTU / T_spreading.
+    Default: 0
+
+tcp_initial_spreading_debug - BOOLEAN
+    When non-zero, a kernel log message will print the rate computed by Initial
+    Spreading each time a TCP connection is initialized.
+    Default: 0
+
 UDP variables:
 
 udp_mem - vector of 3 INTEGERs: min, pressure, max
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 70e55d2..66c6305 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -282,6 +282,8 @@ extern int sysctl_tcp_limit_output_bytes;
 extern int sysctl_tcp_challenge_ack_limit;
 extern unsigned int sysctl_tcp_notsent_lowat;
 extern int sysctl_tcp_min_tso_segs;
+extern int sysctl_tcp_initial_spreading_rate_min;
+extern int sysctl_tcp_initial_spreading_debug;
 
 extern atomic_long_t tcp_memory_allocated;
 extern struct percpu_counter tcp_sockets_allocated;
diff --git a/include/uapi/linux/sysctl.h b/include/uapi/linux/sysctl.h
index 6d67213..355e741 100644
--- a/include/uapi/linux/sysctl.h
+++ b/include/uapi/linux/sysctl.h
@@ -425,6 +425,8 @@ enum
 	NET_TCP_ALLOWED_CONG_CONTROL=123,
 	NET_TCP_MAX_SSTHRESH=124,
 	NET_TCP_FRTO_RESPONSE=125,
+	NET_TCP_INITIAL_SPREADING_RATE_MIN=126,
+	NET_TCP_INITIAL_SPREADING_DEBUG=127,
 };
 
 enum {
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 3d69ec8..47827a4 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -733,6 +733,21 @@ static struct ctl_table ipv4_table[] = {
 		.extra2		= &gso_max_segs,
 	},
 	{
+		.procname	= "tcp_initial_spreading_rate_min",
+		.data		= &sysctl_tcp_initial_spreading_rate_min,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+		.procname	= "tcp_initial_spreading_debug",
+		.data		= &sysctl_tcp_initial_spreading_debug,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec,
+	},
+	{
+
 		.procname	= "udp_mem",
 		.data		= &sysctl_udp_mem,
 		.maxlen		= sizeof(sysctl_udp_mem),
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index c53b7f3..f02f389 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -99,6 +99,9 @@ int sysctl_tcp_thin_dupack __read_mostly;
 int sysctl_tcp_moderate_rcvbuf __read_mostly = 1;
 int sysctl_tcp_early_retrans __read_mostly = 3;
 
+int sysctl_tcp_initial_spreading_rate_min __read_mostly = 0;
+int sysctl_tcp_initial_spreading_debug __read_mostly = 0;
+
 #define FLAG_DATA		0x01 /* Incoming frame contained data.		*/
 #define FLAG_WIN_UPDATE		0x02 /* Incoming ACK was a window update.	*/
 #define FLAG_DATA_ACKED		0x04 /* This ACK acknowledged new data.		*/
@@ -763,6 +766,34 @@ static void tcp_update_pacing_rate(struct sock *sk)
 						sk->sk_max_pacing_rate);
 }
 
+/* Set the sk_pacing_rate to allow FQ packet scheduler doing TCP initial
+ * spreading.
+ * See
+ * https://tools.ietf.org/html/draft-sallantin-tcpm-initial-spreading-00
+ */
+static void tcp_set_initial_pacing_rate(struct sock *sk)
+{
+	const struct tcp_sock *tp = tcp_sk(sk);
+	u64 rate;
+
+	/* set sk_pacing_rate to 100 % of current rate (mss * cwnd / srtt) */
+	rate = (u64)tp->mss_cache * (HZ << 3) * tp->snd_cwnd;
+
+	if (tp->srtt > 8 + 2)
+		do_div(rate, tp->srtt);
+
+    /* rate_min = mtu / T_spreading (see draft) */
+	if (sysctl_tcp_initial_spreading_rate_min)
+		rate = max_t(u64, rate,	sysctl_tcp_initial_spreading_rate_min);
+
+	rate = min_t(u64, rate,	sk->sk_max_pacing_rate);
+
+	if (sysctl_tcp_initial_spreading_debug)
+		printk(KERN_INFO "tcp_pacing_rate: %llu\n", rate);
+
+	ACCESS_ONCE(sk->sk_pacing_rate) = rate;
+}
+
 /* Calculate rto without backoff.  This is the second half of Van Jacobson's
  * routine referred to above.
  */
@@ -5767,7 +5798,7 @@ int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
 		} else
 			tcp_init_metrics(sk);
 
-		tcp_update_pacing_rate(sk);
+		tcp_set_initial_pacing_rate(sk);
 
 		/* Prevent spurious tcp_cwnd_restart() on first data packet */
 		tp->lsndtime = tcp_time_stamp;
diff --git a/net/sched/sch_fq.c b/net/sched/sch_fq.c
index 95d8439..b4063a5 100644
--- a/net/sched/sch_fq.c
+++ b/net/sched/sch_fq.c
@@ -708,8 +708,9 @@ static int fq_init(struct Qdisc *sch, struct nlattr *opt)
 
 	sch->limit		= 10000;
 	q->flow_plimit		= 100;
-	q->quantum		= 2 * psched_mtu(qdisc_dev(sch));
-	q->initial_quantum	= 10 * psched_mtu(qdisc_dev(sch));
+    /* We change default values for initial spreading use-case */
+	q->quantum		= 1 * psched_mtu(qdisc_dev(sch));
+	q->initial_quantum	= 1 * psched_mtu(qdisc_dev(sch));
 	q->flow_refill_delay	= msecs_to_jiffies(40);
 	q->flow_max_rate	= ~0U;
 	q->rate_enable		= 1;
-- 
2.2.2