http://highscalability.com/blog/2010/12/20/netflix-use-less-chatty-protocols-in-the-cloud-plus-26-fixes.html#comment12046108

linux-2.6-el5-net-tcp-iw10.patch

diff -up linux-2.6.18/include/net/tcp.h.orig linux-2.6.18/include/net/tcp.h
--- linux-2.6.18/include/net/tcp.h.orig	2011-09-14 18:58:25.474537285 -0400
+++ linux-2.6.18/include/net/tcp.h	2011-09-14 18:58:37.942381278 -0400
@@ -54,6 +54,9 @@ extern void tcp_time_wait(struct sock *s
  */
 #define MAX_TCP_WINDOW		32767U
 
+/* Offer an initial receive window. */
+#define TCP_DEFAULT_INIT_RCVWND    10
+
 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
 #define TCP_MIN_MSS		88U
 
@@ -185,6 +188,9 @@ extern void tcp_time_wait(struct sock *s
 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
 
+/* TCP initial congestion window */
+#define TCP_INIT_CWND        10
+
 extern struct inet_timewait_death_row tcp_death_row;
 
 /* sysctl variables for tcp */
diff -up linux-2.6.18/net/ipv4/tcp_input.c.orig linux-2.6.18/net/ipv4/tcp_input.c
--- linux-2.6.18/net/ipv4/tcp_input.c.orig	2011-09-14 18:58:46.081626607 -0400
+++ linux-2.6.18/net/ipv4/tcp_input.c	2011-09-14 19:00:26.904278072 -0400
@@ -742,17 +742,12 @@ void tcp_update_metrics(struct sock *sk)
 	}
 }
 
-/* Numbers are taken from RFC2414.  */
 __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst)
 {
 	__u32 cwnd = (dst ? dst_metric(dst, RTAX_INITCWND) : 0);
 
-	if (!cwnd) {
-		if (tp->mss_cache > 1460)
-			cwnd = 2;
-		else
-			cwnd = (tp->mss_cache > 1095) ? 3 : 4;
-	}
+	if (!cwnd)
+		cwnd = TCP_INIT_CWND;
 	return min_t(__u32, cwnd, tp->snd_cwnd_clamp);
 }
 
@@ -834,6 +829,8 @@ static void tcp_init_metrics(struct sock
 	tcp_bound_rto(sk);
 	if (inet_csk(sk)->icsk_rto < TCP_TIMEOUT_INIT && !tp->rx_opt.saw_tstamp)
 		goto reset;
+
+cwnd:
 	tp->snd_cwnd = tcp_init_cwnd(tp, dst);
 	tp->snd_cwnd_stamp = tcp_time_stamp;
 	return;
@@ -848,6 +845,7 @@ reset:
 		tp->mdev = tp->mdev_max = tp->rttvar = TCP_TIMEOUT_INIT;
 		inet_csk(sk)->icsk_rto = TCP_TIMEOUT_INIT;
 	}
+	goto cwnd;
 }
 
 static void tcp_update_reordering(struct sock *sk, const int metric,
diff -up linux-2.6.18/net/ipv4/tcp_output.c.orig linux-2.6.18/net/ipv4/tcp_output.c
--- linux-2.6.18/net/ipv4/tcp_output.c.orig	2011-09-14 19:00:40.465020643 -0400
+++ linux-2.6.18/net/ipv4/tcp_output.c	2011-09-14 19:01:53.911210145 -0400
@@ -208,19 +208,18 @@ void tcp_select_initial_window(int __spa
 		}
 	}
 
-	/* Set initial window to value enough for senders,
-	 * following RFC2414. Senders, not following this RFC,
-	 * will be satisfied with 2.
-	 */
-	if (mss > (1<<*rcv_wscale)) {
-		int init_cwnd = 4;
-		if (mss > 1460*3)
-			init_cwnd = 2;
-		else if (mss > 1460)
-			init_cwnd = 3;
-		if (*rcv_wnd > init_cwnd*mss)
-			*rcv_wnd = init_cwnd*mss;
-	}
+	/* Set initial window to a value enough for senders starting with
+	 * initial congestion window of TCP_DEFAULT_INIT_RCVWND. Place
+	 * a limit on the initial window when mss is larger than 1460.
+	 */
+	if (mss > (1 << *rcv_wscale)) {
+		int init_cwnd = TCP_DEFAULT_INIT_RCVWND;
+		if (mss > 1460)
+			init_cwnd =
+			max_t(u32, (1460 * TCP_DEFAULT_INIT_RCVWND) / mss, 2);
+		else
+			*rcv_wnd = min(*rcv_wnd, init_cwnd * mss);
+    }
 
 	/* Set the clamp no higher than max representable value */
 	(*window_clamp) = min(65535U << (*rcv_wscale), *window_clamp);