[net-next PATCH v5 0/3] net: TCP thin-stream latency-improving modifications

[net-next PATCH v5 0/3] net: TCP thin-stream latency-improving modifications

[Andreas Petlund]

This is a series of patches enabling non-intrusive, dynamically 
triggered modifications that improve retransmission latencies 
for thin streams.

The patch set was modified according to the feedback received.

Major change:
      -Used bitfields to compact the nonagle variable
       in the tcp_sock struct. nonagle, thin_lto and
       thin_dupack is now contained in the same u8.

I decided to use bitfields to handle this as it is
already done similarly in the tcp_options_received struct.

Also corrected some formatting issues.

Cheers,
Andreas Petlund

[net-next PATCH v5 1/3] net: TCP thin-stream detection

[Andreas Petlund]

Inline function to dynamically detect thin streams based on
the number of packets in flight. Used to dynamically trigger
thin-stream mechanisms if enabled by ioctl or sysctl.

Signed-off-by: Andreas Petlund <apetlund@simula.no>
---
 Documentation/networking/tcp-thin.txt |   47 +++++++++++++++++++++++++++++++++
 include/net/tcp.h                     |    8 +++++
 2 files changed, 55 insertions(+), 0 deletions(-)
 create mode 100644 Documentation/networking/tcp-thin.txt

diff --git a/Documentation/networking/tcp-thin.txt b/Documentation/networking/tcp-thin.txt
new file mode 100644
index 0000000..151e229
--- /dev/null
+++ b/Documentation/networking/tcp-thin.txt
@@ -0,0 +1,47 @@
+Thin-streams and TCP
+====================
+A wide range of Internet-based services that use reliable transport
+protocols display what we call thin-stream properties. This means
+that the application sends data with such a low rate that the
+retransmission mechanisms of the transport protocol are not fully
+effective. In time-dependent scenarios (like online games, control
+systems, stock trading etc.) where the user experience depends
+on the data delivery latency, packet loss can be devastating for
+the service quality. Extreme latencies are caused by TCP's
+dependency on the arrival of new data from the application to trigger
+retransmissions effectively through fast retransmit instead of
+waiting for long timeouts.
+
+After analysing a large number of time-dependent interactive
+applications, we have seen that they often produce thin streams
+and also stay with this traffic pattern throughout its entire
+lifespan. The combination of time-dependency and the fact that the
+streams provoke high latencies when using TCP is unfortunate.
+
+In order to reduce application-layer latency when packets are lost,
+a set of mechanisms has been made, which address these latency issues
+for thin streams. In short, if the kernel detects a thin stream,
+the retransmission mechanisms are modified in the following manner:
+
+1) If the stream is thin, fast retransmit on the first dupACK.
+2) If the stream is thin, do not apply exponential backoff.
+
+These enhancements are applied only if the stream is detected as
+thin. This is accomplished by defining a threshold for the number
+of packets in flight. If there are less than 4 packets in flight,
+fast retransmissions can not be triggered, and the stream is prone
+to experience high retransmission latencies.
+
+Since these mechanisms are targeted at time-dependent applications,
+they must be specifically activated by the application using the
+TCP_THIN_LINEAR_TIMEOUTS and TCP_THIN_DUPACK IOCTLS or the
+tcp_thin_linear_timeouts and tcp_thin_dupack sysctls. Both
+modifications are turned off by default.
+
+References
+==========
+More information on the modifications, as well as a wide range of
+experimental data can be found here:
+"Improving latency for interactive, thin-stream applications over
+reliable transport"
+http://simula.no/research/nd/publications/Simula.nd.477/simula_pdf_file
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 75a00c8..0bdc3f6 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -1386,6 +1386,14 @@ static inline void tcp_highest_sack_combine(struct sock *sk,
 		tcp_sk(sk)->highest_sack = new;
 }
 
+/* Determines whether this is a thin stream (which may suffer from
+ * increased latency). Used to trigger latency-reducing mechanisms.
+ */
+static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
+{
+	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
+}
+
 /* /proc */
 enum tcp_seq_states {
 	TCP_SEQ_STATE_LISTENING,
-- 
1.6.3.3

[net-next PATCH v5 2/3] net: TCP thin linear timeouts

[Andreas Petlund]

This patch will make TCP use only linear timeouts if the
stream is thin. This will help to avoid the very high latencies
that thin stream suffer because of exponential backoff. This
mechanism is only active if enabled by iocontrol or syscontrol
and the stream is identified as thin. A maximum of 6 linear
timeouts is tried before exponential backoff is resumed. 

Signed-off-by: Andreas Petlund <apetlund@simula.no>
---
 Documentation/networking/ip-sysctl.txt |   12 ++++++++++++
 include/linux/tcp.h                    |    5 ++++-
 include/net/tcp.h                      |    4 ++++
 net/ipv4/sysctl_net_ipv4.c             |    7 +++++++
 net/ipv4/tcp.c                         |    7 +++++++
 net/ipv4/tcp_timer.c                   |   21 ++++++++++++++++++++-
 6 files changed, 54 insertions(+), 2 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index 2dc7a1d..f147310 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -487,6 +487,18 @@ tcp_dma_copybreak - INTEGER
 	and CONFIG_NET_DMA is enabled.
 	Default: 4096
 
+tcp_thin_linear_timeouts - BOOLEAN
+	Enable dynamic triggering of linear timeouts for thin streams.
+	If set, a check is performed upon retransmission by timeout to
+	determine if the stream is thin (less than 4 packets in flight).
+	As long as the stream is found to be thin, up to 6 linear
+	timeouts may be performed before exponential backoff mode is
+	initiated. This improves retransmission latency for
+	non-aggressive thin streams, often found to be time-dependent.
+	For more information on thin streams, see
+	Documentation/networking/tcp-thin.txt
+	Default: 0
+
 UDP variables:
 
 udp_mem - vector of 3 INTEGERs: min, pressure, max
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 7fee8a4..3ba8b07 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -103,6 +103,7 @@ enum {
 #define TCP_CONGESTION		13	/* Congestion control algorithm */
 #define TCP_MD5SIG		14	/* TCP MD5 Signature (RFC2385) */
 #define TCP_COOKIE_TRANSACTIONS	15	/* TCP Cookie Transactions */
+#define TCP_THIN_LINEAR_TIMEOUTS 16      /* Use linear timeouts for thin streams*/
 
 /* for TCP_INFO socket option */
 #define TCPI_OPT_TIMESTAMPS	1
@@ -340,7 +341,9 @@ struct tcp_sock {
 	u32	frto_highmark;	/* snd_nxt when RTO occurred */
 	u16	advmss;		/* Advertised MSS			*/
 	u8	frto_counter;	/* Number of new acks after RTO */
-	u8	nonagle;	/* Disable Nagle algorithm?             */
+	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
+		thin_lto    : 1,/* Use linear timeouts for thin streams */
+		unused      : 3;
 
 /* RTT measurement */
 	u32	srtt;		/* smoothed round trip time << 3	*/
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 0bdc3f6..6278fc7 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -196,6 +196,9 @@ extern void tcp_time_wait(struct sock *sk, int state, int timeo);
 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
 
+/* TCP thin-stream limits */
+#define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
+
 extern struct inet_timewait_death_row tcp_death_row;
 
 /* sysctl variables for tcp */
@@ -241,6 +244,7 @@ extern int sysctl_tcp_workaround_signed_windows;
 extern int sysctl_tcp_slow_start_after_idle;
 extern int sysctl_tcp_max_ssthresh;
 extern int sysctl_tcp_cookie_size;
+extern int sysctl_tcp_thin_linear_timeouts;
 
 extern atomic_t tcp_memory_allocated;
 extern struct percpu_counter tcp_sockets_allocated;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 7e3712c..e6a2460 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -576,6 +576,13 @@ static struct ctl_table ipv4_table[] = {
 		.proc_handler	= proc_dointvec
 	},
 	{
+		.procname       = "tcp_thin_linear_timeouts",
+		.data           = &sysctl_tcp_thin_linear_timeouts,
+		.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.c b/net/ipv4/tcp.c
index e471d03..21bae9a 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2229,6 +2229,13 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
 		}
 		break;
 
+	case TCP_THIN_LINEAR_TIMEOUTS:
+		if (val < 0 || val > 1)
+			err = -EINVAL;
+		else
+			tp->thin_lto = val;
+		break;
+
 	case TCP_CORK:
 		/* When set indicates to always queue non-full frames.
 		 * Later the user clears this option and we transmit
diff --git a/net/ipv4/tcp_timer.c b/net/ipv4/tcp_timer.c
index de7d1bf..a17629b 100644
--- a/net/ipv4/tcp_timer.c
+++ b/net/ipv4/tcp_timer.c
@@ -29,6 +29,7 @@ int sysctl_tcp_keepalive_intvl __read_mostly = TCP_KEEPALIVE_INTVL;
 int sysctl_tcp_retries1 __read_mostly = TCP_RETR1;
 int sysctl_tcp_retries2 __read_mostly = TCP_RETR2;
 int sysctl_tcp_orphan_retries __read_mostly;
+int sysctl_tcp_thin_linear_timeouts __read_mostly;
 
 static void tcp_write_timer(unsigned long);
 static void tcp_delack_timer(unsigned long);
@@ -415,7 +416,25 @@ void tcp_retransmit_timer(struct sock *sk)
 	icsk->icsk_retransmits++;
 
 out_reset_timer:
-	icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
+	/* If stream is thin, use linear timeouts. Since 'icsk_backoff' is
+	 * used to reset timer, set to 0. Recalculate 'icsk_rto' as this
+	 * might be increased if the stream oscillates between thin and thick,
+	 * thus the old value might already be too high compared to the value
+	 * set by 'tcp_set_rto' in tcp_input.c which resets the rto without
+	 * backoff. Limit to TCP_THIN_LINEAR_RETRIES before initiating
+	 * exponential backoff behaviour to avoid continue hammering
+	 * linear-timeout retransmissions into a black hole
+	 */
+	if (sk->sk_state == TCP_ESTABLISHED &&
+	    (tp->thin_lto || sysctl_tcp_thin_linear_timeouts) &&
+	    tcp_stream_is_thin(tp) &&
+	    icsk->icsk_retransmits <= TCP_THIN_LINEAR_RETRIES) {
+		icsk->icsk_backoff = 0;
+		icsk->icsk_rto = min(__tcp_set_rto(tp), TCP_RTO_MAX);
+	} else {
+		/* Use normal (exponential) backoff */
+		icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
+	}
 	inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto, TCP_RTO_MAX);
 	if (retransmits_timed_out(sk, sysctl_tcp_retries1 + 1))
 		__sk_dst_reset(sk);
-- 
1.6.3.3

[net-next PATCH v5 3/3] net: TCP thin dupack

[Andreas Petlund]

This patch enables fast retransmissions after one dupACK for
TCP if the stream is identified as thin. This will reduce
latencies for thin streams that are not able to trigger fast
retransmissions due to high packet interarrival time. This
mechanism is only active if enabled by iocontrol or syscontrol
and the stream is identified as thin.


Signed-off-by: Andreas Petlund <apetlund@simula.no>
---
 Documentation/networking/ip-sysctl.txt |   12 ++++++++++++
 include/linux/tcp.h                    |    4 +++-
 include/net/tcp.h                      |    1 +
 net/ipv4/sysctl_net_ipv4.c             |    7 +++++++
 net/ipv4/tcp.c                         |    7 +++++++
 net/ipv4/tcp_input.c                   |   12 ++++++++++++
 6 files changed, 42 insertions(+), 1 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index f147310..2571a62 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -499,6 +499,18 @@ tcp_thin_linear_timeouts - BOOLEAN
 	Documentation/networking/tcp-thin.txt
 	Default: 0
 
+tcp_thin_dupack - BOOLEAN
+	Enable dynamic triggering of retransmissions after one dupACK
+	for thin streams. If set, a check is performed upon reception
+	of a dupACK to determine if the stream is thin (less than 4
+	packets in flight). As long as the stream is found to be thin,
+	data is retransmitted on the first received dupACK. This
+	improves retransmission latency for non-aggressive thin
+	streams, often found to be time-dependent.
+	For more information on thin streams, see
+	Documentation/networking/tcp-thin.txt
+	Default: 0
+
 UDP variables:
 
 udp_mem - vector of 3 INTEGERs: min, pressure, max
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 3ba8b07..a778ee0 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -104,6 +104,7 @@ enum {
 #define TCP_MD5SIG		14	/* TCP MD5 Signature (RFC2385) */
 #define TCP_COOKIE_TRANSACTIONS	15	/* TCP Cookie Transactions */
 #define TCP_THIN_LINEAR_TIMEOUTS 16      /* Use linear timeouts for thin streams*/
+#define TCP_THIN_DUPACK         17      /* Fast retrans. after 1 dupack */
 
 /* for TCP_INFO socket option */
 #define TCPI_OPT_TIMESTAMPS	1
@@ -343,7 +344,8 @@ struct tcp_sock {
 	u8	frto_counter;	/* Number of new acks after RTO */
 	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
 		thin_lto    : 1,/* Use linear timeouts for thin streams */
-		unused      : 3;
+		thin_dupack : 1,/* Fast retransmit on first dupack      */
+		unused      : 2;
 
 /* RTT measurement */
 	u32	srtt;		/* smoothed round trip time << 3	*/
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 6278fc7..56f0aec 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -245,6 +245,7 @@ extern int sysctl_tcp_slow_start_after_idle;
 extern int sysctl_tcp_max_ssthresh;
 extern int sysctl_tcp_cookie_size;
 extern int sysctl_tcp_thin_linear_timeouts;
+extern int sysctl_tcp_thin_dupack;
 
 extern atomic_t tcp_memory_allocated;
 extern struct percpu_counter tcp_sockets_allocated;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index e6a2460..c1bc074 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -582,6 +582,13 @@ static struct ctl_table ipv4_table[] = {
 		.mode           = 0644,
 		.proc_handler   = proc_dointvec
 	},
+        {
+		.procname       = "tcp_thin_dupack",
+		.data           = &sysctl_tcp_thin_dupack,
+		.maxlen         = sizeof(int),
+		.mode           = 0644,
+		.proc_handler   = proc_dointvec
+	},
 	{
 		.procname	= "udp_mem",
 		.data		= &sysctl_udp_mem,
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 21bae9a..5901010 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2236,6 +2236,13 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
 			tp->thin_lto = val;
 		break;
 
+	case TCP_THIN_DUPACK:
+		if (val < 0 || val > 1)
+			err = -EINVAL;
+		else
+			tp->thin_dupack = val;
+		break;
+
 	case TCP_CORK:
 		/* When set indicates to always queue non-full frames.
 		 * Later the user clears this option and we transmit
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 3fddc69..8d950b9 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -89,6 +89,8 @@ int sysctl_tcp_frto __read_mostly = 2;
 int sysctl_tcp_frto_response __read_mostly;
 int sysctl_tcp_nometrics_save __read_mostly;
 
+int sysctl_tcp_thin_dupack __read_mostly;
+
 int sysctl_tcp_moderate_rcvbuf __read_mostly = 1;
 int sysctl_tcp_abc __read_mostly;
 
@@ -2447,6 +2449,16 @@ static int tcp_time_to_recover(struct sock *sk)
 		return 1;
 	}
 
+	/* If a thin stream is detected, retransmit after first
+	 * received dupack. Employ only if SACK is supported in order
+	 * to avoid possible corner-case series of spurious retransmissions
+	 * Use only if there are no unsent data.
+	 */
+	if ((tp->thin_dupack || sysctl_tcp_thin_dupack) &&
+	    tcp_stream_is_thin(tp) && tcp_dupack_heuristics(tp) > 1 &&
+	    tcp_is_sack(tp) && sk->sk_send_head == NULL)
+		return 1;
+
 	return 0;
 }
 
-- 
1.6.3.3

Re: [net-next PATCH v5 3/3] net: TCP thin dupack

[Ilpo Järvinen]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 5492 bytes --]

On Thu, 18 Feb 2010, Andreas Petlund wrote:

> This patch enables fast retransmissions after one dupACK for
> TCP if the stream is identified as thin. This will reduce
> latencies for thin streams that are not able to trigger fast
> retransmissions due to high packet interarrival time. This
> mechanism is only active if enabled by iocontrol or syscontrol
> and the stream is identified as thin.
> 
> 
> Signed-off-by: Andreas Petlund <apetlund@simula.no>
> ---
>  Documentation/networking/ip-sysctl.txt |   12 ++++++++++++
>  include/linux/tcp.h                    |    4 +++-
>  include/net/tcp.h                      |    1 +
>  net/ipv4/sysctl_net_ipv4.c             |    7 +++++++
>  net/ipv4/tcp.c                         |    7 +++++++
>  net/ipv4/tcp_input.c                   |   12 ++++++++++++
>  6 files changed, 42 insertions(+), 1 deletions(-)
> 
> diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
> index f147310..2571a62 100644
> --- a/Documentation/networking/ip-sysctl.txt
> +++ b/Documentation/networking/ip-sysctl.txt
> @@ -499,6 +499,18 @@ tcp_thin_linear_timeouts - BOOLEAN
>  	Documentation/networking/tcp-thin.txt
>  	Default: 0
>  
> +tcp_thin_dupack - BOOLEAN
> +	Enable dynamic triggering of retransmissions after one dupACK
> +	for thin streams. If set, a check is performed upon reception
> +	of a dupACK to determine if the stream is thin (less than 4
> +	packets in flight). As long as the stream is found to be thin,
> +	data is retransmitted on the first received dupACK. This
> +	improves retransmission latency for non-aggressive thin
> +	streams, often found to be time-dependent.
> +	For more information on thin streams, see
> +	Documentation/networking/tcp-thin.txt
> +	Default: 0
> +
>  UDP variables:
>  
>  udp_mem - vector of 3 INTEGERs: min, pressure, max
> diff --git a/include/linux/tcp.h b/include/linux/tcp.h
> index 3ba8b07..a778ee0 100644
> --- a/include/linux/tcp.h
> +++ b/include/linux/tcp.h
> @@ -104,6 +104,7 @@ enum {
>  #define TCP_MD5SIG		14	/* TCP MD5 Signature (RFC2385) */
>  #define TCP_COOKIE_TRANSACTIONS	15	/* TCP Cookie Transactions */
>  #define TCP_THIN_LINEAR_TIMEOUTS 16      /* Use linear timeouts for thin streams*/
> +#define TCP_THIN_DUPACK         17      /* Fast retrans. after 1 dupack */
>  
>  /* for TCP_INFO socket option */
>  #define TCPI_OPT_TIMESTAMPS	1
> @@ -343,7 +344,8 @@ struct tcp_sock {
>  	u8	frto_counter;	/* Number of new acks after RTO */
>  	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
>  		thin_lto    : 1,/* Use linear timeouts for thin streams */
> -		unused      : 3;
> +		thin_dupack : 1,/* Fast retransmit on first dupack      */
> +		unused      : 2;
>  
>  /* RTT measurement */
>  	u32	srtt;		/* smoothed round trip time << 3	*/
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index 6278fc7..56f0aec 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -245,6 +245,7 @@ extern int sysctl_tcp_slow_start_after_idle;
>  extern int sysctl_tcp_max_ssthresh;
>  extern int sysctl_tcp_cookie_size;
>  extern int sysctl_tcp_thin_linear_timeouts;
> +extern int sysctl_tcp_thin_dupack;
>  
>  extern atomic_t tcp_memory_allocated;
>  extern struct percpu_counter tcp_sockets_allocated;
> diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
> index e6a2460..c1bc074 100644
> --- a/net/ipv4/sysctl_net_ipv4.c
> +++ b/net/ipv4/sysctl_net_ipv4.c
> @@ -582,6 +582,13 @@ static struct ctl_table ipv4_table[] = {
>  		.mode           = 0644,
>  		.proc_handler   = proc_dointvec
>  	},
> +        {
> +		.procname       = "tcp_thin_dupack",
> +		.data           = &sysctl_tcp_thin_dupack,
> +		.maxlen         = sizeof(int),
> +		.mode           = 0644,
> +		.proc_handler   = proc_dointvec
> +	},
>  	{
>  		.procname	= "udp_mem",
>  		.data		= &sysctl_udp_mem,
> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> index 21bae9a..5901010 100644
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -2236,6 +2236,13 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
>  			tp->thin_lto = val;
>  		break;
>  
> +	case TCP_THIN_DUPACK:
> +		if (val < 0 || val > 1)
> +			err = -EINVAL;
> +		else
> +			tp->thin_dupack = val;
> +		break;
> +
>  	case TCP_CORK:
>  		/* When set indicates to always queue non-full frames.
>  		 * Later the user clears this option and we transmit
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 3fddc69..8d950b9 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -89,6 +89,8 @@ int sysctl_tcp_frto __read_mostly = 2;
>  int sysctl_tcp_frto_response __read_mostly;
>  int sysctl_tcp_nometrics_save __read_mostly;
>  
> +int sysctl_tcp_thin_dupack __read_mostly;
> +
>  int sysctl_tcp_moderate_rcvbuf __read_mostly = 1;
>  int sysctl_tcp_abc __read_mostly;
>  
> @@ -2447,6 +2449,16 @@ static int tcp_time_to_recover(struct sock *sk)
>  		return 1;
>  	}
>  
> +	/* If a thin stream is detected, retransmit after first
> +	 * received dupack. Employ only if SACK is supported in order
> +	 * to avoid possible corner-case series of spurious retransmissions
> +	 * Use only if there are no unsent data.
> +	 */
> +	if ((tp->thin_dupack || sysctl_tcp_thin_dupack) &&
> +	    tcp_stream_is_thin(tp) && tcp_dupack_heuristics(tp) > 1 &&
> +	    tcp_is_sack(tp) && sk->sk_send_head == NULL)

Use tcp_send_head(sk) instead.

> +		return 1;
> +
>  	return 0;
>  }

Other than that,

Acked-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>

-- 
 i.

[net-next PATCH v5.1 3/3] net: TCP thin dupack

[Andreas Petlund]

This patch enables fast retransmissions after one dupACK for
TCP if the stream is identified as thin. This will reduce
latencies for thin streams that are not able to trigger fast
retransmissions due to high packet interarrival time. This
mechanism is only active if enabled by iocontrol or syscontrol
and the stream is identified as thin.

Signed-off-by: Andreas Petlund <apetlund@simula.no>
---
 Documentation/networking/ip-sysctl.txt |   12 ++++++++++++
 include/linux/tcp.h                    |    4 +++-
 include/net/tcp.h                      |    1 +
 net/ipv4/sysctl_net_ipv4.c             |    7 +++++++
 net/ipv4/tcp.c                         |    7 +++++++
 net/ipv4/tcp_input.c                   |   12 ++++++++++++
 6 files changed, 42 insertions(+), 1 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.txt b/Documentation/networking/ip-sysctl.txt
index f147310..2571a62 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -499,6 +499,18 @@ tcp_thin_linear_timeouts - BOOLEAN
 	Documentation/networking/tcp-thin.txt
 	Default: 0
 
+tcp_thin_dupack - BOOLEAN
+	Enable dynamic triggering of retransmissions after one dupACK
+	for thin streams. If set, a check is performed upon reception
+	of a dupACK to determine if the stream is thin (less than 4
+	packets in flight). As long as the stream is found to be thin,
+	data is retransmitted on the first received dupACK. This
+	improves retransmission latency for non-aggressive thin
+	streams, often found to be time-dependent.
+	For more information on thin streams, see
+	Documentation/networking/tcp-thin.txt
+	Default: 0
+
 UDP variables:
 
 udp_mem - vector of 3 INTEGERs: min, pressure, max
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 3ba8b07..a778ee0 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -104,6 +104,7 @@ enum {
 #define TCP_MD5SIG		14	/* TCP MD5 Signature (RFC2385) */
 #define TCP_COOKIE_TRANSACTIONS	15	/* TCP Cookie Transactions */
 #define TCP_THIN_LINEAR_TIMEOUTS 16      /* Use linear timeouts for thin streams*/
+#define TCP_THIN_DUPACK         17      /* Fast retrans. after 1 dupack */
 
 /* for TCP_INFO socket option */
 #define TCPI_OPT_TIMESTAMPS	1
@@ -343,7 +344,8 @@ struct tcp_sock {
 	u8	frto_counter;	/* Number of new acks after RTO */
 	u8	nonagle     : 4,/* Disable Nagle algorithm?             */
 		thin_lto    : 1,/* Use linear timeouts for thin streams */
-		unused      : 3;
+		thin_dupack : 1,/* Fast retransmit on first dupack      */
+		unused      : 2;
 
 /* RTT measurement */
 	u32	srtt;		/* smoothed round trip time << 3	*/
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 6278fc7..56f0aec 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -245,6 +245,7 @@ extern int sysctl_tcp_slow_start_after_idle;
 extern int sysctl_tcp_max_ssthresh;
 extern int sysctl_tcp_cookie_size;
 extern int sysctl_tcp_thin_linear_timeouts;
+extern int sysctl_tcp_thin_dupack;
 
 extern atomic_t tcp_memory_allocated;
 extern struct percpu_counter tcp_sockets_allocated;
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index e6a2460..c1bc074 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -582,6 +582,13 @@ static struct ctl_table ipv4_table[] = {
 		.mode           = 0644,
 		.proc_handler   = proc_dointvec
 	},
+        {
+		.procname       = "tcp_thin_dupack",
+		.data           = &sysctl_tcp_thin_dupack,
+		.maxlen         = sizeof(int),
+		.mode           = 0644,
+		.proc_handler   = proc_dointvec
+	},
 	{
 		.procname	= "udp_mem",
 		.data		= &sysctl_udp_mem,
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 21bae9a..5901010 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2236,6 +2236,13 @@ static int do_tcp_setsockopt(struct sock *sk, int level,
 			tp->thin_lto = val;
 		break;
 
+	case TCP_THIN_DUPACK:
+		if (val < 0 || val > 1)
+			err = -EINVAL;
+		else
+			tp->thin_dupack = val;
+		break;
+
 	case TCP_CORK:
 		/* When set indicates to always queue non-full frames.
 		 * Later the user clears this option and we transmit
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 3fddc69..788851c 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -89,6 +89,8 @@ int sysctl_tcp_frto __read_mostly = 2;
 int sysctl_tcp_frto_response __read_mostly;
 int sysctl_tcp_nometrics_save __read_mostly;
 
+int sysctl_tcp_thin_dupack __read_mostly;
+
 int sysctl_tcp_moderate_rcvbuf __read_mostly = 1;
 int sysctl_tcp_abc __read_mostly;
 
@@ -2447,6 +2449,16 @@ static int tcp_time_to_recover(struct sock *sk)
 		return 1;
 	}
 
+	/* If a thin stream is detected, retransmit after first
+	 * received dupack. Employ only if SACK is supported in order
+	 * to avoid possible corner-case series of spurious retransmissions
+	 * Use only if there are no unsent data.
+	 */
+	if ((tp->thin_dupack || sysctl_tcp_thin_dupack) &&
+	    tcp_stream_is_thin(tp) && tcp_dupack_heuristics(tp) > 1 &&
+	    tcp_is_sack(tp) && !tcp_send_head(sk))
+		return 1;
+
 	return 0;
 }
 
-- 
1.6.3.3

Re: [net-next PATCH v5 0/3] net: TCP thin-stream latency-improving modifications

[David Miller]

From: Andreas Petlund <apetlund@simula.no>
Date: Thu, 18 Feb 2010 13:42:13 +0100

> This is a series of patches enabling non-intrusive, dynamically 
> triggered modifications that improve retransmission latencies 
> for thin streams.
> 
> The patch set was modified according to the feedback received.
> 
> Major change:
>       -Used bitfields to compact the nonagle variable
>        in the tcp_sock struct. nonagle, thin_lto and
>        thin_dupack is now contained in the same u8.
> 
> I decided to use bitfields to handle this as it is
> already done similarly in the tcp_options_received struct.
> 
> Also corrected some formatting issues.

All applied (I made sure the get the updated version of patch #3),
thanks a lot!