[TCP]: Add vegas congestion avoidance support.

A forward port of an old 2.3.x kernel hack done
years ago.  I (DaveM) did the first rough port,
Stephen Hemminger actually cleaned it up and
made it usable.

include/linux/sysctl.h

@@ -327,6 +327,10 @@ enum
 	NET_TCP_WESTWOOD=95,
 	NET_IPV4_IGMP_MAX_MSF=96,
 	NET_TCP_NO_METRICS_SAVE=97,
+	NET_TCP_VEGAS=98,
+	NET_TCP_VEGAS_ALPHA=99,
+	NET_TCP_VEGAS_BETA=100,
+	NET_TCP_VEGAS_GAMMA=101,
 };
 
 enum {

include/linux/tcp.h

@@ -388,6 +388,18 @@ struct tcp_opt {
                 __u32    rtt;
                 __u32    rtt_min;          /* minimum observed RTT */
         } westwood;
+
+/* Vegas variables */
+	struct {
+		__u32	beg_snd_nxt;	/* right edge during last RTT */
+		__u32	beg_snd_una;	/* left edge  during last RTT */
+		__u32	beg_snd_cwnd;	/* saves the size of the cwnd */
+		__u8	do_vegas;	/* do vegas for this connection */
+		__u8	doing_vegas_now;/* if true, do vegas for this RTT */
+		__u16	cntRTT;		/* # of RTTs measured within last RTT */
+		__u32	minRTT;		/* min of RTTs measured within last RTT (in usec) */
+		__u32	baseRTT;	/* the min of all Vegas RTT measurements seen (in usec) */
+	} vegas;
 };
 
 /* WARNING: don't change the layout of the members in tcp_sock! */

include/net/tcp.h

@@ -583,6 +583,10 @@ extern int sysctl_tcp_tw_reuse;
 extern int sysctl_tcp_frto;
 extern int sysctl_tcp_low_latency;
 extern int sysctl_tcp_westwood;
+extern int sysctl_tcp_vegas_cong_avoid;
+extern int sysctl_tcp_vegas_alpha;
+extern int sysctl_tcp_vegas_beta;
+extern int sysctl_tcp_vegas_gamma;
 extern int sysctl_tcp_nometrics_save;
 
 extern atomic_t tcp_memory_allocated;
@@ -1212,8 +1216,56 @@ static inline __u32 tcp_recalc_ssthresh(struct tcp_opt *tp)
 	return max(tp->snd_cwnd >> 1U, 2U);
 }
 
+/* Stop taking Vegas samples for now. */
+#define tcp_vegas_disable(__tp)	((__tp)->vegas.doing_vegas_now = 0)
+
+/* Is this TCP connection using Vegas (regardless of whether it is taking
+ * Vegas measurements at the current time)?
+ */
+#define tcp_is_vegas(__tp)	((__tp)->vegas.do_vegas)
+    
+static inline void tcp_vegas_enable(struct tcp_opt *tp)
+{
+	/* There are several situations when we must "re-start" Vegas:
+	 *
+	 *  o when a connection is established
+	 *  o after an RTO
+	 *  o after fast recovery
+	 *  o when we send a packet and there is no outstanding
+	 *    unacknowledged data (restarting an idle connection)
+	 *
+	 * In these circumstances we cannot do a Vegas calculation at the
+	 * end of the first RTT, because any calculation we do is using
+	 * stale info -- both the saved cwnd and congestion feedback are
+	 * stale.
+	 *
+	 * Instead we must wait until the completion of an RTT during
+	 * which we actually receive ACKs.
+	 */
+    
+	/* Begin taking Vegas samples next time we send something. */
+	tp->vegas.doing_vegas_now = 1;
+     
+	/* Set the beginning of the next send window. */
+	tp->vegas.beg_snd_nxt = tp->snd_nxt;
+
+	tp->vegas.cntRTT = 0;
+	tp->vegas.minRTT = 0x7fffffff;
+}
+
+/* Should we be taking Vegas samples right now? */
+#define tcp_vegas_enabled(__tp)	((__tp)->vegas.doing_vegas_now)
+
+extern void tcp_vegas_init(struct tcp_opt *tp);
+
 static inline void tcp_set_ca_state(struct tcp_opt *tp, u8 ca_state)
 {
+	if (tcp_is_vegas(tp)) {
+		if (ca_state == TCP_CA_Open) 
+			tcp_vegas_enable(tp);
+		else
+			tcp_vegas_disable(tp);
+	}
 	tp->ca_state = ca_state;
 }
 

net/ipv4/sysctl_net_ipv4.c

@@ -609,7 +609,38 @@ ctl_table ipv4_table[] = {
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec,
 	},
-
+	{
+		.ctl_name	= NET_TCP_VEGAS,
+		.procname	= "tcp_vegas_cong_avoid",
+		.data		= &sysctl_tcp_vegas_cong_avoid,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
+	{
+		.ctl_name	= NET_TCP_VEGAS_ALPHA,
+		.procname	= "tcp_vegas_alpha",
+		.data		= &sysctl_tcp_vegas_alpha,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
+	{
+		.ctl_name	= NET_TCP_VEGAS_BETA,
+		.procname	= "tcp_vegas_beta",
+		.data		= &sysctl_tcp_vegas_beta,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
+	{
+		.ctl_name	= NET_TCP_VEGAS_GAMMA,
+		.procname	= "tcp_vegas_gamma",
+		.data		= &sysctl_tcp_vegas_gamma,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec,
+	},
 	{ .ctl_name = 0 }
 };
 

net/ipv4/tcp_input.c

@@ -86,8 +86,17 @@ int sysctl_tcp_stdurg;
 int sysctl_tcp_rfc1337;
 int sysctl_tcp_max_orphans = NR_FILE;
 int sysctl_tcp_frto;
-int sysctl_tcp_westwood;
 int sysctl_tcp_nometrics_save;
+int sysctl_tcp_westwood;
+int sysctl_tcp_vegas_cong_avoid;
+
+/* Default values of the Vegas variables, in fixed-point representation
+ * with V_PARAM_SHIFT bits to the right of the binary point.
+ */
+#define V_PARAM_SHIFT 1
+int sysctl_tcp_vegas_alpha = 1<<V_PARAM_SHIFT;
+int sysctl_tcp_vegas_beta  = 3<<V_PARAM_SHIFT;
+int sysctl_tcp_vegas_gamma = 1<<V_PARAM_SHIFT;
 
 #define FLAG_DATA		0x01 /* Incoming frame contained data.		*/
 #define FLAG_WIN_UPDATE		0x02 /* Incoming ACK was a window update.	*/
@@ -404,6 +413,42 @@ static void tcp_event_data_recv(struct sock *sk, struct tcp_opt *tp, struct sk_b
 		tcp_grow_window(sk, tp, skb);
 }
 
+/* Set up a new TCP connection, depending on whether it should be
+ * using Vegas or not.
+ */    
+void tcp_vegas_init(struct tcp_opt *tp)
+{
+	if (sysctl_tcp_vegas_cong_avoid) {
+		tp->vegas.do_vegas = 1;
+		tp->vegas.baseRTT = 0x7fffffff;
+		tcp_vegas_enable(tp);
+	} else 
+		tcp_vegas_disable(tp);
+}
+
+/* Do RTT sampling needed for Vegas.
+ * Basically we:
+ *   o min-filter RTT samples from within an RTT to get the current
+ *     propagation delay + queuing delay (we are min-filtering to try to
+ *     avoid the effects of delayed ACKs)
+ *   o min-filter RTT samples from a much longer window (forever for now)
+ *     to find the propagation delay (baseRTT)
+ */
+static inline void vegas_rtt_calc(struct tcp_opt *tp, __u32 rtt)
+{
+	__u32 vrtt = rtt + 1; /* Never allow zero rtt or baseRTT */
+
+	/* Filter to find propagation delay: */
+	if (vrtt < tp->vegas.baseRTT) 
+		tp->vegas.baseRTT = vrtt;
+
+	/* Find the min RTT during the last RTT to find
+	 * the current prop. delay + queuing delay:
+	 */
+	tp->vegas.minRTT = min(tp->vegas.minRTT, vrtt);
+	tp->vegas.cntRTT++;
+}
+
 /* Called to compute a smoothed rtt estimate. The data fed to this
  * routine either comes from timestamps, or from segments that were
  * known _not_ to have been retransmitted [see Karn/Partridge
@@ -417,6 +462,9 @@ static void tcp_rtt_estimator(struct tcp_opt *tp, __u32 mrtt)
 {
 	long m = mrtt; /* RTT */
 
+	if (tcp_vegas_enabled(tp))
+		vegas_rtt_calc(tp, mrtt);
+
 	/*	The following amusing code comes from Jacobson's
 	 *	article in SIGCOMM '88.  Note that rtt and mdev
 	 *	are scaled versions of rtt and mean deviation.
@@ -1810,11 +1858,10 @@ tcp_ack_update_rtt(struct tcp_opt *tp, int flag, s32 seq_rtt)
 	else if (seq_rtt >= 0)
 		tcp_ack_no_tstamp(tp, seq_rtt, flag);
 }
-
 /* This is Jacobson's slow start and congestion avoidance. 
  * SIGCOMM '88, p. 328.
  */
-static __inline__ void tcp_cong_avoid(struct tcp_opt *tp)
+static __inline__ void reno_cong_avoid(struct tcp_opt *tp)
 {
         if (tp->snd_cwnd <= tp->snd_ssthresh) {
                 /* In "safe" area, increase. */
@@ -1834,6 +1881,236 @@ static __inline__ void tcp_cong_avoid(struct tcp_opt *tp)
 	tp->snd_cwnd_stamp = tcp_time_stamp;
 }
 
+/* This is based on the congestion detection/avoidance scheme described in
+ *    Lawrence S. Brakmo and Larry L. Peterson.
+ *    "TCP Vegas: End to end congestion avoidance on a global internet."
+ *    IEEE Journal on Selected Areas in Communication, 13(8):1465--1480,
+ *    October 1995. Available from:
+ *	ftp://ftp.cs.arizona.edu/xkernel/Papers/jsac.ps
+ *
+ * See http://www.cs.arizona.edu/xkernel/ for their implementation.
+ * The main aspects that distinguish this implementation from the
+ * Arizona Vegas implementation are:
+ *   o We do not change the loss detection or recovery mechanisms of
+ *     Linux in any way. Linux already recovers from losses quite well,
+ *     using fine-grained timers, NewReno, and FACK.
+ *   o To avoid the performance penalty imposed by increasing cwnd
+ *     only every-other RTT during slow start, we increase during
+ *     every RTT during slow start, just like Reno.
+ *   o Largely to allow continuous cwnd growth during slow start,
+ *     we use the rate at which ACKs come back as the "actual"
+ *     rate, rather than the rate at which data is sent.
+ *   o To speed convergence to the right rate, we set the cwnd
+ *     to achieve the right ("actual") rate when we exit slow start.
+ *   o To filter out the noise caused by delayed ACKs, we use the
+ *     minimum RTT sample observed during the last RTT to calculate
+ *     the actual rate.
+ *   o When the sender re-starts from idle, it waits until it has
+ *     received ACKs for an entire flight of new data before making
+ *     a cwnd adjustment decision. The original Vegas implementation
+ *     assumed senders never went idle.
+ */
+static void vegas_cong_avoid(struct tcp_opt *tp, u32 ack, u32 seq_rtt)
+{
+	/* The key players are v_beg_snd_una and v_beg_snd_nxt.
+	 *
+	 * These are so named because they represent the approximate values
+	 * of snd_una and snd_nxt at the beginning of the current RTT. More
+	 * precisely, they represent the amount of data sent during the RTT.
+	 * At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
+	 * we will calculate that (v_beg_snd_nxt - v_beg_snd_una) outstanding
+	 * bytes of data have been ACKed during the course of the RTT, giving
+	 * an "actual" rate of:
+	 *
+	 *     (v_beg_snd_nxt - v_beg_snd_una) / (rtt duration)
+	 *
+	 * Unfortunately, v_beg_snd_una is not exactly equal to snd_una,
+	 * because delayed ACKs can cover more than one segment, so they
+	 * don't line up nicely with the boundaries of RTTs.
+	 *
+	 * Another unfortunate fact of life is that delayed ACKs delay the
+	 * advance of the left edge of our send window, so that the number
+	 * of bytes we send in an RTT is often less than our cwnd will allow.
+	 * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
+	 */
+
+	if (after(ack, tp->vegas.beg_snd_nxt)) {
+		/* Do the Vegas once-per-RTT cwnd adjustment. */
+		u32 old_wnd, old_snd_cwnd;
+
+		
+		/* Here old_wnd is essentially the window of data that was
+		 * sent during the previous RTT, and has all
+		 * been acknowledged in the course of the RTT that ended
+		 * with the ACK we just received. Likewise, old_snd_cwnd
+		 * is the cwnd during the previous RTT.
+		 */
+		old_wnd = (tp->vegas.beg_snd_nxt - tp->vegas.beg_snd_una) /
+			tp->mss_cache;
+		old_snd_cwnd = tp->vegas.beg_snd_cwnd;
+
+		/* Save the extent of the current window so we can use this
+		 * at the end of the next RTT.
+		 */
+		tp->vegas.beg_snd_una  = tp->vegas.beg_snd_nxt;
+		tp->vegas.beg_snd_nxt  = tp->snd_nxt;
+		tp->vegas.beg_snd_cwnd = tp->snd_cwnd;
+
+		/* Take into account the current RTT sample too, to
+		 * decrease the impact of delayed acks. This double counts
+		 * this sample since we count it for the next window as well,
+		 * but that's not too awful, since we're taking the min,
+		 * rather than averaging.
+		 */
+		vegas_rtt_calc(tp, seq_rtt);
+
+		/* We do the Vegas calculations only if we got enough RTT
+		 * samples that we can be reasonably sure that we got
+		 * at least one RTT sample that wasn't from a delayed ACK.
+		 * If we only had 2 samples total,
+		 * then that means we're getting only 1 ACK per RTT, which
+		 * means they're almost certainly delayed ACKs.
+		 * If  we have 3 samples, we should be OK.
+		 */
+
+		if (tp->vegas.cntRTT <= 2) {
+			/* We don't have enough RTT samples to do the Vegas
+			 * calculation, so we'll behave like Reno.
+			 */
+			if (tp->snd_cwnd > tp->snd_ssthresh)
+				tp->snd_cwnd++;
+		} else {
+			u32 rtt, target_cwnd, diff;
+
+			/* We have enough RTT samples, so, using the Vegas
+			 * algorithm, we determine if we should increase or
+			 * decrease cwnd, and by how much.
+			 */
+
+			/* Pluck out the RTT we are using for the Vegas
+			 * calculations. This is the min RTT seen during the
+			 * last RTT. Taking the min filters out the effects
+			 * of delayed ACKs, at the cost of noticing congestion
+			 * a bit later.
+			 */
+			rtt = tp->vegas.minRTT;
+
+			/* Calculate the cwnd we should have, if we weren't
+			 * going too fast.
+			 *
+			 * This is:
+			 *     (actual rate in segments) * baseRTT
+			 * We keep it as a fixed point number with
+			 * V_PARAM_SHIFT bits to the right of the binary point.
+			 */
+			target_cwnd = ((old_wnd * tp->vegas.baseRTT)
+				       << V_PARAM_SHIFT) / rtt;
+
+			/* Calculate the difference between the window we had,
+			 * and the window we would like to have. This quantity
+			 * is the "Diff" from the Arizona Vegas papers.
+			 *
+			 * Again, this is a fixed point number with
+			 * V_PARAM_SHIFT bits to the right of the binary
+			 * point.
+			 */
+			diff = (old_wnd << V_PARAM_SHIFT) - target_cwnd;
+
+			if (tp->snd_cwnd < tp->snd_ssthresh) {
+				/* Slow start.  */
+				if (diff > sysctl_tcp_vegas_gamma) {
+					/* Going too fast. Time to slow down
+					 * and switch to congestion avoidance.
+					 */
+					tp->snd_ssthresh = 2;
+
+					/* Set cwnd to match the actual rate
+					 * exactly:
+					 *   cwnd = (actual rate) * baseRTT
+					 * Then we add 1 because the integer
+					 * truncation robs us of full link
+					 * utilization.
+					 */
+					tp->snd_cwnd = min(tp->snd_cwnd,
+							   (target_cwnd >>
+							    V_PARAM_SHIFT)+1);
+
+				}
+			} else {
+				/* Congestion avoidance. */
+				u32 next_snd_cwnd;
+
+				/* Figure out where we would like cwnd
+				 * to be.
+				 */
+				if (diff > sysctl_tcp_vegas_beta) {
+					/* The old window was too fast, so
+					 * we slow down.
+					 */
+					next_snd_cwnd = old_snd_cwnd - 1;
+				} else if (diff < sysctl_tcp_vegas_alpha) {
+					/* We don't have enough extra packets
+					 * in the network, so speed up.
+					 */
+					next_snd_cwnd = old_snd_cwnd + 1;
+				} else {
+					/* Sending just as fast as we
+					 * should be.
+					 */
+					next_snd_cwnd = old_snd_cwnd;
+				}
+
+				/* Adjust cwnd upward or downward, toward the
+				 * desired value.
+				 */
+				if (next_snd_cwnd > tp->snd_cwnd)
+					tp->snd_cwnd++;
+				else if (next_snd_cwnd < tp->snd_cwnd)
+					tp->snd_cwnd--;
+			}
+		}
+
+		/* Wipe the slate clean for the next RTT. */
+		tp->vegas.cntRTT = 0;
+		tp->vegas.minRTT = 0x7fffffff;
+	}
+
+	/* The following code is executed for every ack we receive,
+	 * except for conditions checked in should_advance_cwnd()
+	 * before the call to tcp_cong_avoid(). Mainly this means that
+	 * we only execute this code if the ack actually acked some
+	 * data.
+	 */
+
+	/* If we are in slow start, increase our cwnd in response to this ACK.
+	 * (If we are not in slow start then we are in congestion avoidance,
+	 * and adjust our congestion window only once per RTT. See the code
+	 * above.)
+	 */
+	if (tp->snd_cwnd <= tp->snd_ssthresh) 
+		tp->snd_cwnd++;
+
+	/* to keep cwnd from growing without bound */
+	tp->snd_cwnd = min_t(u32, tp->snd_cwnd, tp->snd_cwnd_clamp);
+
+	/* Make sure that we are never so timid as to reduce our cwnd below
+	 * 2 MSS.
+	 *
+	 * Going below 2 MSS would risk huge delayed ACKs from our receiver.
+	 */
+	tp->snd_cwnd = max(tp->snd_cwnd, 2U);
+
+	tp->snd_cwnd_stamp = tcp_time_stamp;
+}
+
+static inline void tcp_cong_avoid(struct tcp_opt *tp, u32 ack, u32 seq_rtt)
+{
+	if (tcp_vegas_enabled(tp))
+		vegas_cong_avoid(tp, ack, seq_rtt);
+	else
+		reno_cong_avoid(tp);
+}
+
 /* Restart timer after forward progress on connection.
  * RFC2988 recommends to restart timer to now+rto.
  */
@@ -1848,7 +2125,7 @@ static __inline__ void tcp_ack_packets_out(struct sock *sk, struct tcp_opt *tp)
 }
 
 /* Remove acknowledged frames from the retransmission queue. */
-static int tcp_clean_rtx_queue(struct sock *sk)
+static int tcp_clean_rtx_queue(struct sock *sk, __s32 *seq_rtt_p)
 {
 	struct tcp_opt *tp = tcp_sk(sk);
 	struct sk_buff *skb;
@@ -1934,6 +2211,7 @@ static int tcp_clean_rtx_queue(struct sock *sk)
 		}
 	}
 #endif
+	*seq_rtt_p = seq_rtt;
 	return acked;
 }
 
@@ -2302,6 +2580,7 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
 	u32 ack_seq = TCP_SKB_CB(skb)->seq;
 	u32 ack = TCP_SKB_CB(skb)->ack_seq;
 	u32 prior_in_flight;
+	s32 seq_rtt;
 	int prior_packets;
 
 	/* If the ack is newer than sent or older than previous acks
@@ -2353,7 +2632,7 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
 	prior_in_flight = tcp_packets_in_flight(tp);
 
 	/* See if we can take anything off of the retransmit queue. */
-	flag |= tcp_clean_rtx_queue(sk);
+	flag |= tcp_clean_rtx_queue(sk, &seq_rtt);
 
 	if (tp->frto_counter)
 		tcp_process_frto(sk, prior_snd_una);
@@ -2361,13 +2640,14 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
 	if (tcp_ack_is_dubious(tp, flag)) {
 		/* Advanve CWND, if state allows this. */
 		if ((flag & FLAG_DATA_ACKED) &&
-		    prior_in_flight >= tp->snd_cwnd &&
+		    (tcp_vegas_enabled(tp) || prior_in_flight >= tp->snd_cwnd) &&
 		    tcp_may_raise_cwnd(tp, flag))
-			tcp_cong_avoid(tp);
+			tcp_cong_avoid(tp, ack, seq_rtt);
 		tcp_fastretrans_alert(sk, prior_snd_una, prior_packets, flag);
 	} else {
-		if ((flag & FLAG_DATA_ACKED) && prior_in_flight >= tp->snd_cwnd)
-			tcp_cong_avoid(tp);
+		if ((flag & FLAG_DATA_ACKED) && 
+		    (tcp_vegas_enabled(tp) || prior_in_flight >= tp->snd_cwnd))
+			tcp_cong_avoid(tp, ack, seq_rtt);
 	}
 
 	if ((flag & FLAG_FORWARD_PROGRESS) || !(flag&FLAG_NOT_DUP))

net/ipv4/tcp_minisocks.c

@@ -841,6 +841,7 @@ struct sock *tcp_create_openreq_child(struct sock *sk, struct open_request *req,
 		if (newtp->ecn_flags&TCP_ECN_OK)
 			newsk->sk_no_largesend = 1;
 
+		tcp_vegas_init(newtp);
 		TCP_INC_STATS_BH(TcpPassiveOpens);
 	}
 	return newsk;

net/ipv4/tcp_output.c

@@ -106,6 +106,9 @@ static void tcp_cwnd_restart(struct tcp_opt *tp, struct dst_entry *dst)
 	u32 restart_cwnd = tcp_init_cwnd(tp, dst);
 	u32 cwnd = tp->snd_cwnd;
 
+	if (tcp_is_vegas(tp)) 
+		tcp_vegas_enable(tp);
+
 	tp->snd_ssthresh = tcp_current_ssthresh(tp);
 	restart_cwnd = min(restart_cwnd, cwnd);
 
@@ -225,6 +228,19 @@ int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb)
 			tcp_header_size += (TCPOLEN_SACK_BASE_ALIGNED +
 					    (tp->eff_sacks * TCPOLEN_SACK_PERBLOCK));
 		}
+		
+		/*
+		 * If the connection is idle and we are restarting,
+		 * then we don't want to do any Vegas calculations
+		 * until we get fresh RTT samples.  So when we
+		 * restart, we reset our Vegas state to a clean
+		 * slate. After we get acks for this flight of
+		 * packets, _then_ we can make Vegas calculations
+		 * again.
+		 */
+		if (tcp_is_vegas(tp) && tcp_packets_in_flight(tp) == 0)
+			tcp_vegas_enable(tp);
+
 		th = (struct tcphdr *) skb_push(skb, tcp_header_size);
 		skb->h.th = th;
 		skb_set_owner_w(skb, sk);
@@ -1268,6 +1284,7 @@ static inline void tcp_connect_init(struct sock *sk)
 		tp->window_clamp = dst_metric(dst, RTAX_WINDOW);
 	tp->advmss = dst_metric(dst, RTAX_ADVMSS);
 	tcp_initialize_rcv_mss(sk);
+	tcp_vegas_init(tp);
 
 	tcp_select_initial_window(tcp_full_space(sk),
 				  tp->advmss - (tp->ts_recent_stamp ? tp->tcp_header_len - sizeof(struct tcphdr) : 0),
@@ -1318,6 +1335,7 @@ int tcp_connect(struct sock *sk)
 	TCP_SKB_CB(buff)->end_seq = tp->write_seq;
 	tp->snd_nxt = tp->write_seq;
 	tp->pushed_seq = tp->write_seq;
+	tcp_vegas_init(tp);
 
 	/* Send it off. */
 	TCP_SKB_CB(buff)->when = tcp_time_stamp;