[PATCH] net/ipv4, linux-2.6.30.4

[PATCH] net/ipv4, linux-2.6.30.4

[Daniel Slot]

[-- Attachment #1: Type: text/plain, Size: 938 bytes --]

RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
In the absence of explicit congestion notification from the network,
TCP uses loss as an indication of congestion.
One of the ways TCP detects loss is using the arrival of three
duplicate acknowledgments.
However, this heuristic is not always correct, notably in the case
when network paths reorder segments (for whatever reason), resulting
in degraded performance.
TCP-NCR is designed to mitigate this degraded performance by
increasing the number of duplicate acknowledgments required to trigger
loss recovery,
based on the current state of the connection, in an effort to better
disambiguate true segment loss from segment reordering.
This document specifies the changes to TCP, as well as the costs and
benefits of these modifications.

This patch adds TCP-NCR as socket option to the Linux kernel (version 2.6.30.4).
Written by Daniel Slot, Email: slot.daniel(at)gmail.com

[-- Attachment #2: README.txt --]
[-- Type: text/plain, Size: 234 bytes --]

This patch adds TCP-NCR as socket option to the Linux kernel.
To use TCP-NCR in careful mode (resp. aggressive mode),
an application has to set the TCP-NCR socket option (23) to the value 1 (resp. 2) when it starts a TCP connection.

[-- Attachment #3: patch-linuxkernel-2.6.30-tcp_ncr --]
[-- Type: application/octet-stream, Size: 6904 bytes --]

diff -uprN linux-2.6.30.4/include/linux/tcp.h linux-2.6.30.4-NCR/include/linux/tcp.h
--- /include/linux/tcp.h	2009-07-31 00:34:47.000000000 +0200
+++ /include/linux/tcp.h	2009-08-12 20:15:18.000000000 +0200
@@ -96,6 +96,7 @@ enum { 
 #define TCP_QUICKACK		12	/* Block/reenable quick acks */
 #define TCP_CONGESTION		13	/* Congestion control algorithm */
 #define TCP_MD5SIG		14	/* TCP MD5 Signature (RFC2385) */
+#define TCP_NCR         23  /* TCP NCR (RFC4653) */
 
 #define TCPI_OPT_TIMESTAMPS	1
 #define TCPI_OPT_SACK		2
@@ -408,6 +409,13 @@ struct tcp_sock {
 #endif
 
 	int			linger2;
+
+/* TCP NCR extension information */
+    u8  tcp_ncr_flag;
+    u8  elt_flag;
+    u8  dupthresh;
+    u8  LT_F;
+    u32 priorFlightSize;
 };
 
 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
diff -uprN linux-2.6.30.4/net/ipv4/tcp.c linux-2.6.30.4-NCR/net/ipv4/tcp.c
--- /net/ipv4/tcp.c	2009-07-31 00:34:47.000000000 +0200
+++ /net/ipv4/tcp.c	2009-08-12 20:15:18.000000000 +0200
@@ -2208,6 +2208,17 @@ static int do_tcp_setsockopt(struct sock
 		break;
 #endif
 
+    case TCP_NCR:
+        /* TCP-NCR : val equal 1 for careful mode, val equal 2 for aggressive mode */
+        if (val){
+            tp->tcp_ncr_flag = 1;
+            if (val==1) tp->LT_F = 3;
+            if (val==2) tp->LT_F = 4;
+        } else {
+            tp->tcp_ncr_flag = 0;
+        }
+        break;
+
 	default:
 		err = -ENOPROTOOPT;
 		break;
diff -uprN linux-2.6.30.4/net/ipv4/tcp_input.c linux-2.6.30.4-NCR/net/ipv4/tcp_input.c
--- /net/ipv4/tcp_input.c	2009-07-31 00:34:47.000000000 +0200
+++ /net/ipv4/tcp_input.c	2009-08-12 20:15:18.000000000 +0200
@@ -1003,6 +1003,45 @@ static void tcp_skb_mark_lost_uncond_ver
 	}
 }
 
+/* TCP-NCR: Test if TCP-NCR may be used
+ * (Following RFC 4653 recommendations)
+ */
+static int tcp_ncr_test(struct tcp_sock *tp)
+{
+    return (tp->tcp_ncr_flag && tcp_is_sack(tp) && !(tp->nonagle & TCP_NAGLE_OFF));
+}
+
+/* TCP-NCR: Initiate Extended Limited Transmit
+ * (RFC 4653 Initialization)
+ * */
+static void tcp_ncr_elt_init(struct tcp_sock *tp, int how)
+{
+    if (!how) tp->priorFlightSize = tp->packets_out;
+    tp->elt_flag = 1;
+    tp->dupthresh = max_t(u32, ((2 * tp->packets_out)/tp->LT_F), 3);
+}
+
+/* TCP-NCR Extended Limited Transmit
+ * (RFC 4653 Termination)
+ */
+static void tcp_ncr_elt_end(struct tcp_sock *tp, int flag , int how)
+{
+    if (how){
+        /* New cumulative ACK during ELT, it is reordering. */
+        tp->snd_ssthresh = tp->priorFlightSize;
+        tp->snd_cwnd = min(tp->packets_out+1, tp->priorFlightSize);
+        tp->snd_cwnd_stamp = tcp_time_stamp;
+        if (flag & FLAG_DATA_SACKED) tcp_ncr_elt_init(tp, 1);
+        else tp->elt_flag = 0;
+    } else {
+        /* Dupthresh is reached, start recovery */
+        tp->snd_ssthresh = (tp->priorFlightSize/2);
+        tp->snd_cwnd = tp->snd_ssthresh;
+        tp->snd_cwnd_stamp = tcp_time_stamp;
+        tp->elt_flag = 0;
+    }
+}
+
 /* This procedure tags the retransmission queue when SACKs arrive.
  *
  * We have three tag bits: SACKED(S), RETRANS(R) and LOST(L).
@@ -1346,6 +1385,9 @@ static u8 tcp_sacktag_one(struct sk_buff
 			}
 		}
 
+        /* TCP-NCR: Initialization */
+        if (tcp_ncr_test(tp) && (!tp->elt_flag) && (tp->sacked_out == 0)) tcp_ncr_elt_init(tp, 0);
+
 		sacked |= TCPCB_SACKED_ACKED;
 		state->flag |= FLAG_DATA_SACKED;
 		tp->sacked_out += pcount;
@@ -2425,9 +2467,13 @@ static int tcp_time_to_recover(struct so
 	if (tp->lost_out)
 		return 1;
 
-	/* Not-A-Trick#2 : Classic rule... */
-	if (tcp_dupack_heurestics(tp) > tp->reordering)
-		return 1;
+    /* Not-A-Trick#2 : Classic rule...
+     * (Option to use TCP-NCR dupthresh instead)
+     */
+    if (tp->elt_flag && (tcp_dupack_heurestics(tp) > tp->dupthresh))
+        return 1;
+    if (!tp->elt_flag && (tcp_dupack_heurestics(tp) > tp->reordering))
+        return 1;
 
 	/* Trick#3 : when we use RFC2988 timer restart, fast
 	 * retransmit can be triggered by timeout of queue head.
@@ -2603,6 +2649,17 @@ static void tcp_cwnd_down(struct sock *s
 	}
 }
 
+/* TCP-NCR: Extended Limited Transmit
+ * (RFC 4653 Main Part)
+ */
+static void tcp_ncr_elt(struct sock *sk, int flag)
+{
+    struct tcp_sock *tp = tcp_sk(sk);
+
+    if (tp->LT_F == 3) tcp_cwnd_down(sk, flag);
+    tp->dupthresh = max_t(u32, ((2 * tp->packets_out)/tp->LT_F), 3);
+}
+
 /* Nothing was retransmitted or returned timestamp is less
  * than timestamp of the first retransmission.
  */
@@ -2812,7 +2869,7 @@ static void tcp_try_to_open(struct sock 
 
 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) {
 		tcp_try_keep_open(sk);
-		tcp_moderate_cwnd(tp);
+        if (!tcp_ncr_test(tp)) tcp_moderate_cwnd(tp);
 	} else {
 		tcp_cwnd_down(sk, flag);
 	}
@@ -2920,6 +2977,9 @@ static void tcp_fastretrans_alert(struct
 	if (WARN_ON(!tp->sacked_out && tp->fackets_out))
 		tp->fackets_out = 0;
 
+    /* TCP-NCR: Extended Limited Transmit */
+    if (tp->elt_flag && (flag & FLAG_DATA_SACKED)) tcp_ncr_elt(sk, flag);
+
 	/* Now state machine starts.
 	 * A. ECE, hence prohibit cwnd undoing, the reduction is required. */
 	if (flag & FLAG_ECE)
@@ -3050,7 +3110,8 @@ static void tcp_fastretrans_alert(struct
 		if (icsk->icsk_ca_state < TCP_CA_CWR) {
 			if (!(flag & FLAG_ECE))
 				tp->prior_ssthresh = tcp_current_ssthresh(sk);
-			tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
+            if (tp->elt_flag) tcp_ncr_elt_end(tp, flag, 0);
+            else tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
 			TCP_ECN_queue_cwr(tp);
 		}
 
@@ -3062,8 +3123,8 @@ static void tcp_fastretrans_alert(struct
 
 	if (do_lost || (tcp_is_fack(tp) && tcp_head_timedout(sk)))
 		tcp_update_scoreboard(sk, fast_rexmit);
-	tcp_cwnd_down(sk, flag);
-	tcp_xmit_retransmit_queue(sk);
+    if (!tcp_ncr_test(tp))tcp_cwnd_down(sk, flag);
+    tcp_xmit_retransmit_queue(sk);
 }
 
 static void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
@@ -3285,8 +3346,10 @@ static int tcp_clean_rtx_queue(struct so
 			int delta;
 
 			/* Non-retransmitted hole got filled? That's reordering */
-			if (reord < prior_fackets)
+			if (reord < prior_fackets){
 				tcp_update_reordering(sk, tp->fackets_out - reord, 0);
+                if (tp->elt_flag) tcp_ncr_elt_end(tp, flag, 1);
+            }
 
 			delta = tcp_is_fack(tp) ? pkts_acked :
 						  prior_sacked - tp->sacked_out;
diff -uprN linux-2.6.30.4/net/ipv4/tcp_ipv4.c linux-2.6.30.4-NCR/net/ipv4/tcp_ipv4.c
--- /net/ipv4/tcp_ipv4.c	2009-07-31 00:34:47.000000000 +0200
+++ /net/ipv4/tcp_ipv4.c	2009-08-12 20:15:18.000000000 +0200
@@ -1774,6 +1774,11 @@ static int tcp_v4_init_sock(struct sock 
 	tp->mss_cache = 536;
 
 	tp->reordering = sysctl_tcp_reordering;
+
+    /* TCP-NCR: Initiate some variables */
+    tp->dupthresh = TCP_FASTRETRANS_THRESH;
+    tp->elt_flag = 0;
+
 	icsk->icsk_ca_ops = &tcp_init_congestion_ops;
 
 	sk->sk_state = TCP_CLOSE;

[PATCH] net/ipv4, linux-2.6.30.4

[Daniel Slot]

RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
In the absence of explicit congestion notification from the network,
TCP uses loss as an indication of congestion.
One of the ways TCP detects loss is using the arrival of three
duplicate acknowledgments.
However, this heuristic is not always correct, notably in the case
when network paths reorder segments (for whatever reason), resulting
in degraded performance.
TCP-NCR is designed to mitigate this degraded performance by
increasing the number of duplicate acknowledgments required to trigger
loss recovery,
based on the current state of the connection, in an effort to better
disambiguate true segment loss from segment reordering.
This document specifies the changes to TCP, as well as the costs and
benefits of these modifications.

This patch adds TCP-NCR as socket option to the Linux kernel (version 2.6.30.4).
To use TCP-NCR in careful mode (resp. aggressive mode),
an application has to set the TCP-NCR socket option (23) to the value
1 (resp. 2) \
when it starts a TCP connection.

Written by Daniel Slot, Email: slot.daniel(at)gmail.com

---------------------------

diff -uprN linux-2.6.30.4/include/linux/tcp.h
linux-2.6.30.4-NCR/include/linux/tcp.h
--- /include/linux/tcp.h	2009-07-31 00:34:47.000000000 +0200
+++ /include/linux/tcp.h	2009-08-12 20:15:18.000000000 +0200
@@ -96,6 +96,7 @@ enum {
 #define TCP_QUICKACK		12	/* Block/reenable quick acks */
 #define TCP_CONGESTION		13	/* Congestion control algorithm */
 #define TCP_MD5SIG		14	/* TCP MD5 Signature (RFC2385) */
+#define TCP_NCR         23  /* TCP NCR (RFC4653) */

 #define TCPI_OPT_TIMESTAMPS	1
 #define TCPI_OPT_SACK		2
@@ -408,6 +409,13 @@ struct tcp_sock {
 #endif

 	int			linger2;
+
+/* TCP NCR extension information */
+    u8  tcp_ncr_flag;
+    u8  elt_flag;
+    u8  dupthresh;
+    u8  LT_F;
+    u32 priorFlightSize;
 };

 static inline struct tcp_sock *tcp_sk(const struct sock *sk)
diff -uprN linux-2.6.30.4/net/ipv4/tcp.c linux-2.6.30.4-NCR/net/ipv4/tcp.c
--- /net/ipv4/tcp.c	2009-07-31 00:34:47.000000000 +0200
+++ /net/ipv4/tcp.c	2009-08-12 20:15:18.000000000 +0200
@@ -2208,6 +2208,17 @@ static int do_tcp_setsockopt(struct sock
 		break;
 #endif

+    case TCP_NCR:
+        /* TCP-NCR : val equal 1 for careful mode, val equal 2 for
aggressive mode */
+        if (val){
+            tp->tcp_ncr_flag = 1;
+            if (val==1) tp->LT_F = 3;
+            if (val==2) tp->LT_F = 4;
+        } else {
+            tp->tcp_ncr_flag = 0;
+        }
+        break;
+
 	default:
 		err = -ENOPROTOOPT;
 		break;
diff -uprN linux-2.6.30.4/net/ipv4/tcp_input.c
linux-2.6.30.4-NCR/net/ipv4/tcp_input.c
--- /net/ipv4/tcp_input.c	2009-07-31 00:34:47.000000000 +0200
+++ /net/ipv4/tcp_input.c	2009-08-12 20:15:18.000000000 +0200
@@ -1003,6 +1003,45 @@ static void tcp_skb_mark_lost_uncond_ver
 	}
 }

+/* TCP-NCR: Test if TCP-NCR may be used
+ * (Following RFC 4653 recommendations)
+ */
+static int tcp_ncr_test(struct tcp_sock *tp)
+{
+    return (tp->tcp_ncr_flag && tcp_is_sack(tp) && !(tp->nonagle &
TCP_NAGLE_OFF));
+}
+
+/* TCP-NCR: Initiate Extended Limited Transmit
+ * (RFC 4653 Initialization)
+ * */
+static void tcp_ncr_elt_init(struct tcp_sock *tp, int how)
+{
+    if (!how) tp->priorFlightSize = tp->packets_out;
+    tp->elt_flag = 1;
+    tp->dupthresh = max_t(u32, ((2 * tp->packets_out)/tp->LT_F), 3);
+}
+
+/* TCP-NCR Extended Limited Transmit
+ * (RFC 4653 Termination)
+ */
+static void tcp_ncr_elt_end(struct tcp_sock *tp, int flag , int how)
+{
+    if (how){
+        /* New cumulative ACK during ELT, it is reordering. */
+        tp->snd_ssthresh = tp->priorFlightSize;
+        tp->snd_cwnd = min(tp->packets_out+1, tp->priorFlightSize);
+        tp->snd_cwnd_stamp = tcp_time_stamp;
+        if (flag & FLAG_DATA_SACKED) tcp_ncr_elt_init(tp, 1);
+        else tp->elt_flag = 0;
+    } else {
+        /* Dupthresh is reached, start recovery */
+        tp->snd_ssthresh = (tp->priorFlightSize/2);
+        tp->snd_cwnd = tp->snd_ssthresh;
+        tp->snd_cwnd_stamp = tcp_time_stamp;
+        tp->elt_flag = 0;
+    }
+}
+
 /* This procedure tags the retransmission queue when SACKs arrive.
  *
  * We have three tag bits: SACKED(S), RETRANS(R) and LOST(L).
@@ -1346,6 +1385,9 @@ static u8 tcp_sacktag_one(struct sk_buff
 			}
 		}

+        /* TCP-NCR: Initialization */
+        if (tcp_ncr_test(tp) && (!tp->elt_flag) && (tp->sacked_out ==
0)) tcp_ncr_elt_init(tp, 0);
+
 		sacked |= TCPCB_SACKED_ACKED;
 		state->flag |= FLAG_DATA_SACKED;
 		tp->sacked_out += pcount;
@@ -2425,9 +2467,13 @@ static int tcp_time_to_recover(struct so
 	if (tp->lost_out)
 		return 1;

-	/* Not-A-Trick#2 : Classic rule... */
-	if (tcp_dupack_heurestics(tp) > tp->reordering)
-		return 1;
+    /* Not-A-Trick#2 : Classic rule...
+     * (Option to use TCP-NCR dupthresh instead)
+     */
+    if (tp->elt_flag && (tcp_dupack_heurestics(tp) > tp->dupthresh))
+        return 1;
+    if (!tp->elt_flag && (tcp_dupack_heurestics(tp) > tp->reordering))
+        return 1;

 	/* Trick#3 : when we use RFC2988 timer restart, fast
 	 * retransmit can be triggered by timeout of queue head.
@@ -2603,6 +2649,17 @@ static void tcp_cwnd_down(struct sock *s
 	}
 }

+/* TCP-NCR: Extended Limited Transmit
+ * (RFC 4653 Main Part)
+ */
+static void tcp_ncr_elt(struct sock *sk, int flag)
+{
+    struct tcp_sock *tp = tcp_sk(sk);
+
+    if (tp->LT_F == 3) tcp_cwnd_down(sk, flag);
+    tp->dupthresh = max_t(u32, ((2 * tp->packets_out)/tp->LT_F), 3);
+}
+
 /* Nothing was retransmitted or returned timestamp is less
  * than timestamp of the first retransmission.
  */
@@ -2812,7 +2869,7 @@ static void tcp_try_to_open(struct sock

 	if (inet_csk(sk)->icsk_ca_state != TCP_CA_CWR) {
 		tcp_try_keep_open(sk);
-		tcp_moderate_cwnd(tp);
+        if (!tcp_ncr_test(tp)) tcp_moderate_cwnd(tp);
 	} else {
 		tcp_cwnd_down(sk, flag);
 	}
@@ -2920,6 +2977,9 @@ static void tcp_fastretrans_alert(struct
 	if (WARN_ON(!tp->sacked_out && tp->fackets_out))
 		tp->fackets_out = 0;

+    /* TCP-NCR: Extended Limited Transmit */
+    if (tp->elt_flag && (flag & FLAG_DATA_SACKED)) tcp_ncr_elt(sk, flag);
+
 	/* Now state machine starts.
 	 * A. ECE, hence prohibit cwnd undoing, the reduction is required. */
 	if (flag & FLAG_ECE)
@@ -3050,7 +3110,8 @@ static void tcp_fastretrans_alert(struct
 		if (icsk->icsk_ca_state < TCP_CA_CWR) {
 			if (!(flag & FLAG_ECE))
 				tp->prior_ssthresh = tcp_current_ssthresh(sk);
-			tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
+            if (tp->elt_flag) tcp_ncr_elt_end(tp, flag, 0);
+            else tp->snd_ssthresh = icsk->icsk_ca_ops->ssthresh(sk);
 			TCP_ECN_queue_cwr(tp);
 		}

@@ -3062,8 +3123,8 @@ static void tcp_fastretrans_alert(struct

 	if (do_lost || (tcp_is_fack(tp) && tcp_head_timedout(sk)))
 		tcp_update_scoreboard(sk, fast_rexmit);
-	tcp_cwnd_down(sk, flag);
-	tcp_xmit_retransmit_queue(sk);
+    if (!tcp_ncr_test(tp))tcp_cwnd_down(sk, flag);
+    tcp_xmit_retransmit_queue(sk);
 }

 static void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt)
@@ -3285,8 +3346,10 @@ static int tcp_clean_rtx_queue(struct so
 			int delta;

 			/* Non-retransmitted hole got filled? That's reordering */
-			if (reord < prior_fackets)
+			if (reord < prior_fackets){
 				tcp_update_reordering(sk, tp->fackets_out - reord, 0);
+                if (tp->elt_flag) tcp_ncr_elt_end(tp, flag, 1);
+            }

 			delta = tcp_is_fack(tp) ? pkts_acked :
 						  prior_sacked - tp->sacked_out;
diff -uprN linux-2.6.30.4/net/ipv4/tcp_ipv4.c
linux-2.6.30.4-NCR/net/ipv4/tcp_ipv4.c
--- /net/ipv4/tcp_ipv4.c	2009-07-31 00:34:47.000000000 +0200
+++ /net/ipv4/tcp_ipv4.c	2009-08-12 20:15:18.000000000 +0200
@@ -1774,6 +1774,11 @@ static int tcp_v4_init_sock(struct sock
 	tp->mss_cache = 536;

 	tp->reordering = sysctl_tcp_reordering;
+
+    /* TCP-NCR: Initiate some variables */
+    tp->dupthresh = TCP_FASTRETRANS_THRESH;
+    tp->elt_flag = 0;
+
 	icsk->icsk_ca_ops = &tcp_init_congestion_ops;

 	sk->sk_state = TCP_CLOSE;
---------------------------

Re: [PATCH] net/ipv4, linux-2.6.30.4

[Stephen Hemminger]

On Wed, 12 Aug 2009 20:50:59 +0200
Daniel Slot <slot.daniel@gmail.com> wrote:

> RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
> In the absence of explicit congestion notification from the network,
> TCP uses loss as an indication of congestion.
> One of the ways TCP detects loss is using the arrival of three
> duplicate acknowledgments.
> However, this heuristic is not always correct, notably in the case
> when network paths reorder segments (for whatever reason), resulting
> in degraded performance.
> TCP-NCR is designed to mitigate this degraded performance by
> increasing the number of duplicate acknowledgments required to trigger
> loss recovery,
> based on the current state of the connection, in an effort to better
> disambiguate true segment loss from segment reordering.
> This document specifies the changes to TCP, as well as the costs and
> benefits of these modifications.
> 
> This patch adds TCP-NCR as socket option to the Linux kernel (version 2.6.30.4).
> Written by Daniel Slot, Email: slot.daniel(at)gmail.com

Patch has funny indentation and awkward naming for socket elements.
Your style needs to match existing code.

What is the usage model for this? I expect that some user who wants
to enable this would be stuck somewhere with a lossy network and
would want to enable it. Or is it something only researchers will
want to play with?

It would be easier to use a sysctl value for this because otherwise
each application has to be changed to select the socket option.

-- 

Re: [PATCH] net/ipv4, linux-2.6.30.4

[Daniel Slot]

I already tried to adapt the style to existing code.
Would be nice if you could give me hint about what is akward.
I'm quite new to the kernelhacking area and sorry for all obvious errors.

The usage is imho interesting in research domains.
As it is an implementation of an ietf RFC, there might be some
researchers who can use it.
It is part of my master thesis and I use it for measurements and comparisons.

I tried to avoid a sysctl value for this.
I don't wanted this algorithm to be used by all TCP connections.

2009/8/12 Stephen Hemminger <shemminger@vyatta.com>:
> On Wed, 12 Aug 2009 20:50:59 +0200
> Daniel Slot <slot.daniel@gmail.com> wrote:
>
>> RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
>> In the absence of explicit congestion notification from the network,
>> TCP uses loss as an indication of congestion.
>> One of the ways TCP detects loss is using the arrival of three
>> duplicate acknowledgments.
>> However, this heuristic is not always correct, notably in the case
>> when network paths reorder segments (for whatever reason), resulting
>> in degraded performance.
>> TCP-NCR is designed to mitigate this degraded performance by
>> increasing the number of duplicate acknowledgments required to trigger
>> loss recovery,
>> based on the current state of the connection, in an effort to better
>> disambiguate true segment loss from segment reordering.
>> This document specifies the changes to TCP, as well as the costs and
>> benefits of these modifications.
>>
>> This patch adds TCP-NCR as socket option to the Linux kernel (version 2.6.30.4).
>> Written by Daniel Slot, Email: slot.daniel(at)gmail.com
>
> Patch has funny indentation and awkward naming for socket elements.
> Your style needs to match existing code.
>
> What is the usage model for this? I expect that some user who wants
> to enable this would be stuck somewhere with a lossy network and
> would want to enable it. Or is it something only researchers will
> want to play with?
>
> It would be easier to use a sysctl value for this because otherwise
> each application has to be changed to select the socket option.
>
> --
>

Re: [PATCH] net/ipv4, linux-2.6.30.4

[David Miller]

From: Daniel Slot <slot.daniel@gmail.com>
Date: Wed, 12 Aug 2009 20:47:44 +0200

> RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
> In the absence of explicit congestion notification from the network, TCP
> uses loss as an indication of congestion.
> One of the ways TCP detects loss is using the arrival of three duplicate
> acknowledgments.
> However, this heuristic is not always correct,
> notably in the case when network paths reorder segments (for whatever
> reason), resulting in degraded performance.

Linux's TCP stack already has sophisticated reordering detection.

> TCP-NCR is designed to mitigate this degraded performance by increasing the
> number of duplicate acknowledgments required to trigger loss recovery,
> based on the current state of the connection, in an effort to better
> disambiguate true segment loss from segment reordering.

We already have code in the stack which tries to detect packet
reordering with a high level of sophistication.

Re: [PATCH] net/ipv4, linux-2.6.30.4

[Arnd Hannemann]

David Miller schrieb:
> From: Daniel Slot <slot.daniel@gmail.com>
> Date: Wed, 12 Aug 2009 20:47:44 +0200
> 
>> RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
>> In the absence of explicit congestion notification from the network, TCP
>> uses loss as an indication of congestion.
>> One of the ways TCP detects loss is using the arrival of three duplicate
>> acknowledgments.
>> However, this heuristic is not always correct,
>> notably in the case when network paths reorder segments (for whatever
>> reason), resulting in degraded performance.
> 
> Linux's TCP stack already has sophisticated reordering detection.ä


Hmm, sophisticated? Sorry, it seemed pretty rudimental/random to me.

Firstly, tp->reordering never shrinks for a given connection
unless an RTO occurs. If that happens tp->reordering is reset to sysctl_tcp_reordering
(but it was initialized with a potentially different value from destination cache).
Why?

Secondly, it simply disables FACK? Disabling FACK completely may (or not) be
the correct solution, if reordering is present. But why don't reenable FACK
after no more reordering is detected? It won't even get re-enabled if an RTO occurs.
It seems even more strange that tp->reordering is used in FACK paths, too.
So if one sets a high sysctl_tcp_reordering, because one expects reordering,
tcp_update_reordering will probably NOT disable FACK, but instead FACK will
be used with a high tp->reordering value.

Thirdly, in most cases it will only
trigger if spurious retransmits already happened. If it triggers in advance (due to
SACK logic, the updated reordering metric will be IMO one to small, leading again
to a spurios retransmit, if a reordering event with the same length will happen again)
IOW it will mostly only reduce the damage to congestion control, but will send out
spurious packets nevertheless.

In my point of view, on should at least build some EWMA or histogram, or build some
whatever statistcs to measure detected reordering and based on this measurement,
adjust the dupthresh (or max_burst, or whatever). Off course, there is always
the question of how much better such an sophisticated statistic will work,
than the current very pragmatic solution...

Please correct me if I'm wrong or just too stupid to understand this stuff.
(very likely;-)

> 
>> TCP-NCR is designed to mitigate this degraded performance by increasing the
>> number of duplicate acknowledgments required to trigger loss recovery,
>> based on the current state of the connection, in an effort to better
>> disambiguate true segment loss from segment reordering.
> 
> We already have code in the stack which tries to detect packet
> reordering with a high level of sophistication.

On the contrary RFC 4653 does not even try to detect reordering. It simply
delays the congestion response in a way which seems very straightforward.
Of course there is the negative impact of increased latency. (Loss recovery
takes longer). However, for large ftp/http transfers, who cares about latency?

There must be some logic in the kernel to detect applications which are
doing bulk transfers for the buffer autotuning, what about enabling RFC 4653
in case such an application is detected?

Daniel, I would assume RFC 4653 would simply work with FACK, at least if
there is no reordering present?

Best regards,
Arnd


-- 
Dipl.-Inform. Arnd Hannemann
RWTH Aachen University
Dept. of Computer Science, Informatik 4
Ahornstr. 55, D-52074 Aachen, Germany
Phone: (+49 241) 80-21423 Fax: (+49 241) 80-22220

Re: [PATCH] net/ipv4, linux-2.6.30.4

[David Miller]

From: Arnd Hannemann <hannemann@nets.rwth-aachen.de>
Date: Thu, 13 Aug 2009 14:40:42 +0200

> David Miller schrieb:
>> From: Daniel Slot <slot.daniel@gmail.com>
>> Date: Wed, 12 Aug 2009 20:47:44 +0200
>> 
>>> RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP.
>>> In the absence of explicit congestion notification from the network, TCP
>>> uses loss as an indication of congestion.
>>> One of the ways TCP detects loss is using the arrival of three duplicate
>>> acknowledgments.
>>> However, this heuristic is not always correct,
>>> notably in the case when network paths reorder segments (for whatever
>>> reason), resulting in degraded performance.
>> 
>> Linux's TCP stack already has sophisticated reordering detection.ä
> 
> 
> Hmm, sophisticated? Sorry, it seemed pretty rudimental/random to me.

It's more sophisticated than what you're proposing, which is a hammer
which requires intervention on the part of the application.

Re: [PATCH] net/ipv4, linux-2.6.30.4

[David Miller]

From: Arnd Hannemann <hannemann@nets.rwth-aachen.de>
Date: Thu, 13 Aug 2009 14:40:42 +0200

> Please correct me if I'm wrong or just too stupid to understand this stuff.
> (very likely;-)

You may be right, but at least the infrastructure for reordering
detection is there.

So instead of improving the stuff we have, to make it detect
when reordering goes away and potentially re-enable FACK, you're
simply adding something new to replace it.

Why not just fix up for the deficiencies you're aware of?

Re: [PATCH] net/ipv4, linux-2.6.30.4

[Daniel Slot]

> Why not just fix up for the deficiencies you're aware of?

Because it never was our intention to replace/improve the already
existing reordering mechanisms.
We just wanted to provide the possibility to switch to another
reordering-robust algorithm.
It could be similar to the large amount of congestion control
algorithms, which are also part of the kernel but rarely used.
A sophistcated congestion control algorithm is standardly used by the
kernel, but other ones are available too.

With our tcp-ncr patch, researchers would have the possibility to make
measurements with an IETF approved algorithm.