[TCP] YeAH-TCP: algorithm implementation

YeAH-TCP is a sender-side high-speed enabled TCP congestion control
algorithm, which uses a mixed loss/delay approach to compute the
congestion window. It's design goals target high efficiency, internal,
RTT and Reno fairness, resilience to link loss while keeping network
elements load as low as possible.

For further details look here:
    http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf

Signed-off-by: Angelo P. Castellani <angelo.castellani@gmail.con>
Signed-off-by: David S. Miller <davem@davemloft.net>

4 files changed, 437 insertions, 0 deletions

diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig
index 9e8ef509c51d..dc61e6641624 100644
--- a/net/ipv4/Kconfig
+++ b/net/ipv4/Kconfig
@@ -574,6 +574,20 @@ config TCP_CONG_VENO
 	loss packets.
 	See http://www.ntu.edu.sg/home5/ZHOU0022/papers/CPFu03a.pdf
 
+config TCP_CONG_YEAH
+	tristate "YeAH TCP"
+	depends on EXPERIMENTAL
+	default n
+	---help---
+	YeAH-TCP is a sender-side high-speed enabled TCP congestion control
+	algorithm, which uses a mixed loss/delay approach to compute the
+	congestion window. It's design goals target high efficiency,
+	internal, RTT and Reno fairness, resilience to link loss while
+	keeping network elements load as low as possible.
+
+	For further details look here:
+	  http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf
+
 choice
 	prompt "Default TCP congestion control"
 	default DEFAULT_CUBIC
diff --git a/net/ipv4/Makefile b/net/ipv4/Makefile
index 7a068626feea..eeb94d5cac96 100644
--- a/net/ipv4/Makefile
+++ b/net/ipv4/Makefile
@@ -49,6 +49,7 @@ obj-$(CONFIG_TCP_CONG_VEGAS) += tcp_vegas.o
 obj-$(CONFIG_TCP_CONG_VENO) += tcp_veno.o
 obj-$(CONFIG_TCP_CONG_SCALABLE) += tcp_scalable.o
 obj-$(CONFIG_TCP_CONG_LP) += tcp_lp.o
+obj-$(CONFIG_TCP_CONG_YEAH) += tcp_yeah.o
 obj-$(CONFIG_NETLABEL) += cipso_ipv4.o
 
 obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \
diff --git a/net/ipv4/tcp_yeah.c b/net/ipv4/tcp_yeah.c
new file mode 100644
index 000000000000..815e020e98fe
--- /dev/null
+++ b/net/ipv4/tcp_yeah.c
@@ -0,0 +1,288 @@
+/*
+ *
+ *   YeAH TCP
+ *
+ * For further details look at:
+ *    http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf
+ *
+ */
+
+#include "tcp_yeah.h"
+
+/* 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
+
+#define TCP_YEAH_ALPHA       80 //lin number of packets queued at the bottleneck
+#define TCP_YEAH_GAMMA        1 //lin fraction of queue to be removed per rtt
+#define TCP_YEAH_DELTA        3 //log minimum fraction of cwnd to be removed on loss
+#define TCP_YEAH_EPSILON      1 //log maximum fraction to be removed on early decongestion
+#define TCP_YEAH_PHY          8 //lin maximum delta from base
+#define TCP_YEAH_RHO         16 //lin minumum number of consecutive rtt to consider competition on loss
+#define TCP_YEAH_ZETA        50 //lin minimum number of state switchs to reset reno_count
+
+#define TCP_SCALABLE_AI_CNT	 100U
+
+/* YeAH variables */
+struct yeah {
+	/* Vegas */
+	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	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) */
+
+	/* YeAH */
+	u32 lastQ;
+	u32 doing_reno_now;
+
+	u32 reno_count;
+	u32 fast_count;
+
+	u32 pkts_acked;
+};
+
+static void tcp_yeah_init(struct sock *sk)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+
+	tcp_vegas_init(sk);
+
+	yeah->doing_reno_now = 0;
+	yeah->lastQ = 0;
+
+	yeah->reno_count = 2;
+
+	/* Ensure the MD arithmetic works.  This is somewhat pedantic,
+	 * since I don't think we will see a cwnd this large. :) */
+	tp->snd_cwnd_clamp = min_t(u32, tp->snd_cwnd_clamp, 0xffffffff/128);
+
+}
+
+
+static void tcp_yeah_pkts_acked(struct sock *sk, u32 pkts_acked)
+{
+	const struct inet_connection_sock *icsk = inet_csk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+
+	if (icsk->icsk_ca_state == TCP_CA_Open)
+		yeah->pkts_acked = pkts_acked;
+}
+
+/* 64bit divisor, dividend and result. dynamic precision */
+static inline u64 div64_64(u64 dividend, u64 divisor)
+{
+	u32 d = divisor;
+
+	if (divisor > 0xffffffffULL) {
+		unsigned int shift = fls(divisor >> 32);
+
+		d = divisor >> shift;
+		dividend >>= shift;
+	}
+
+	/* avoid 64 bit division if possible */
+	if (dividend >> 32)
+		do_div(dividend, d);
+	else
+		dividend = (u32) dividend / d;
+
+	return dividend;
+}
+
+static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack,
+				 u32 seq_rtt, u32 in_flight, int flag)
+{
+	struct tcp_sock *tp = tcp_sk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+
+	if (!tcp_is_cwnd_limited(sk, in_flight))
+		return;
+
+	if (tp->snd_cwnd <= tp->snd_ssthresh) {
+		tcp_slow_start(tp);
+	} else if (!yeah->doing_reno_now) {
+		/* Scalable */
+
+		tp->snd_cwnd_cnt+=yeah->pkts_acked;
+		if (tp->snd_cwnd_cnt > min(tp->snd_cwnd, TCP_SCALABLE_AI_CNT)){
+			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
+				tp->snd_cwnd++;
+			tp->snd_cwnd_cnt = 0;
+		}
+
+		yeah->pkts_acked = 1;
+
+	} else {
+		/* Reno */
+
+		if (tp->snd_cwnd_cnt < tp->snd_cwnd)
+			tp->snd_cwnd_cnt++;
+
+		if (tp->snd_cwnd_cnt >= tp->snd_cwnd) {
+			tp->snd_cwnd++;
+			tp->snd_cwnd_cnt = 0;
+		}
+	}
+
+	/* 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 yeahly 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, yeah->beg_snd_nxt)) {
+
+		/* 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 (yeah->cntRTT > 2) {
+			u32 rtt;
+			u32 queue, maxqueue;
+
+			/* 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 = yeah->minRTT;
+
+			queue = (u32)div64_64((u64)tp->snd_cwnd * (rtt - yeah->baseRTT), rtt);
+
+			maxqueue = TCP_YEAH_ALPHA;
+
+			if (queue > maxqueue ||
+				    rtt - yeah->baseRTT > (yeah->baseRTT / TCP_YEAH_PHY)) {
+
+				if (queue > maxqueue && tp->snd_cwnd > yeah->reno_count) {
+					u32 reduction = min( queue / TCP_YEAH_GAMMA ,
+					                 tp->snd_cwnd >> TCP_YEAH_EPSILON );
+
+					tp->snd_cwnd -= reduction;
+
+					tp->snd_cwnd = max( tp->snd_cwnd, yeah->reno_count);
+
+					tp->snd_ssthresh = tp->snd_cwnd;
+			}
+
+				if (yeah->reno_count <= 2)
+					yeah->reno_count = max( tp->snd_cwnd>>1, 2U);
+				else
+					yeah->reno_count++;
+
+				yeah->doing_reno_now =
+					           min_t( u32, yeah->doing_reno_now + 1 , 0xffffff);
+
+			} else {
+				yeah->fast_count++;
+
+				if (yeah->fast_count > TCP_YEAH_ZETA) {
+					yeah->reno_count = 2;
+					yeah->fast_count = 0;
+				}
+
+				yeah->doing_reno_now = 0;
+			}
+
+			yeah->lastQ = queue;
+
+		}
+
+		/* Save the extent of the current window so we can use this
+		 * at the end of the next RTT.
+		 */
+		yeah->beg_snd_una  = yeah->beg_snd_nxt;
+		yeah->beg_snd_nxt  = tp->snd_nxt;
+		yeah->beg_snd_cwnd = tp->snd_cwnd;
+
+		/* Wipe the slate clean for the next RTT. */
+		yeah->cntRTT = 0;
+		yeah->minRTT = 0x7fffffff;
+	}
+}
+
+static u32 tcp_yeah_ssthresh(struct sock *sk) {
+	const struct tcp_sock *tp = tcp_sk(sk);
+	struct yeah *yeah = inet_csk_ca(sk);
+	u32 reduction;
+
+	if (yeah->doing_reno_now < TCP_YEAH_RHO) {
+		reduction = yeah->lastQ;
+
+		reduction = min( reduction, max(tp->snd_cwnd>>1, 2U) );
+
+		reduction = max( reduction, tp->snd_cwnd >> TCP_YEAH_DELTA);
+	} else
+		reduction = max(tp->snd_cwnd>>1,2U);
+
+	yeah->fast_count = 0;
+	yeah->reno_count = max(yeah->reno_count>>1, 2U);
+
+	return tp->snd_cwnd - reduction;
+}
+
+static struct tcp_congestion_ops tcp_yeah = {
+	.init		= tcp_yeah_init,
+	.ssthresh	= tcp_yeah_ssthresh,
+	.cong_avoid	= tcp_yeah_cong_avoid,
+	.min_cwnd	= tcp_reno_min_cwnd,
+	.rtt_sample	= tcp_vegas_rtt_calc,
+	.set_state	= tcp_vegas_state,
+	.cwnd_event	= tcp_vegas_cwnd_event,
+	.get_info	= tcp_vegas_get_info,
+	.pkts_acked	= tcp_yeah_pkts_acked,
+
+	.owner		= THIS_MODULE,
+	.name		= "yeah",
+};
+
+static int __init tcp_yeah_register(void)
+{
+	BUG_ON(sizeof(struct yeah) > ICSK_CA_PRIV_SIZE);
+	tcp_register_congestion_control(&tcp_yeah);
+	return 0;
+}
+
+static void __exit tcp_yeah_unregister(void)
+{
+	tcp_unregister_congestion_control(&tcp_yeah);
+}
+
+module_init(tcp_yeah_register);
+module_exit(tcp_yeah_unregister);
+
+MODULE_AUTHOR("Angelo P. Castellani");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("YeAH TCP");
diff --git a/net/ipv4/tcp_yeah.h b/net/ipv4/tcp_yeah.h
new file mode 100644
index 000000000000..b3255dba4e2d
--- /dev/null
+++ b/net/ipv4/tcp_yeah.h
@@ -0,0 +1,134 @@
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/skbuff.h>
+#include <linux/inet_diag.h>
+
+#include <net/tcp.h>
+
+/* Vegas variables */
+struct vegas {
+	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	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) */
+};
+
+/* 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.
+ */
+static inline void vegas_enable(struct sock *sk)
+{
+	const struct tcp_sock *tp = tcp_sk(sk);
+	struct vegas *vegas = inet_csk_ca(sk);
+
+	/* Begin taking Vegas samples next time we send something. */
+	vegas->doing_vegas_now = 1;
+
+	/* Set the beginning of the next send window. */
+	vegas->beg_snd_nxt = tp->snd_nxt;
+
+	vegas->cntRTT = 0;
+	vegas->minRTT = 0x7fffffff;
+}
+
+/* Stop taking Vegas samples for now. */
+static inline void vegas_disable(struct sock *sk)
+{
+	struct vegas *vegas = inet_csk_ca(sk);
+
+	vegas->doing_vegas_now = 0;
+}
+
+static void tcp_vegas_init(struct sock *sk)
+{
+	struct vegas *vegas = inet_csk_ca(sk);
+
+	vegas->baseRTT = 0x7fffffff;
+	vegas_enable(sk);
+}
+
+static void tcp_vegas_state(struct sock *sk, u8 ca_state)
+{
+
+	if (ca_state == TCP_CA_Open)
+		vegas_enable(sk);
+	else
+		vegas_disable(sk);
+}
+
+/* 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 void tcp_vegas_rtt_calc(struct sock *sk, u32 usrtt)
+{
+	struct vegas *vegas = inet_csk_ca(sk);
+	u32 vrtt = usrtt + 1; /* Never allow zero rtt or baseRTT */
+
+	/* Filter to find propagation delay: */
+	if (vrtt < vegas->baseRTT)
+		vegas->baseRTT = vrtt;
+
+	/* Find the min RTT during the last RTT to find
+	 * the current prop. delay + queuing delay:
+	 */
+	vegas->minRTT = min(vegas->minRTT, vrtt);
+	vegas->cntRTT++;
+}
+
+/*
+ * 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.
+ */
+static void tcp_vegas_cwnd_event(struct sock *sk, enum tcp_ca_event event)
+{
+	if (event == CA_EVENT_CWND_RESTART ||
+	    event == CA_EVENT_TX_START)
+		tcp_vegas_init(sk);
+}
+
+/* Extract info for Tcp socket info provided via netlink. */
+static void tcp_vegas_get_info(struct sock *sk, u32 ext,
+			       struct sk_buff *skb)
+{
+	const struct vegas *ca = inet_csk_ca(sk);
+	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
+		struct tcpvegas_info *info;
+
+		info = RTA_DATA(__RTA_PUT(skb, INET_DIAG_VEGASINFO,
+					  sizeof(*info)));
+
+		info->tcpv_enabled = ca->doing_vegas_now;
+		info->tcpv_rttcnt = ca->cntRTT;
+		info->tcpv_rtt = ca->baseRTT;
+		info->tcpv_minrtt = ca->minRTT;
+	rtattr_failure:	;
+	}
+}
+
+