[linux.git] / net / ipv4 / tcp_bic.c

/*
 * Binary Increase Congestion control for TCP
 *
 * This is from the implementation of BICTCP in
 * Lison-Xu, Kahaled Harfoush, and Injong Rhee.
 *  "Binary Increase Congestion Control for Fast, Long Distance
 *  Networks" in InfoComm 2004
 * Available from:
 *  http://www.csc.ncsu.edu/faculty/rhee/export/bitcp.pdf
 *
 * Unless BIC is enabled and congestion window is large
 * this behaves the same as the original Reno.
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <net/tcp.h>


#define BICTCP_BETA_SCALE    1024	/* Scale factor beta calculation
					 * max_cwnd = snd_cwnd * beta
					 */
#define BICTCP_B		4	 /*
					  * In binary search,
					  * go to point (max+min)/N
					  */

static int fast_convergence = 1;
static int max_increment = 16;
static int low_window = 14;
static int beta = 819;		/* = 819/1024 (BICTCP_BETA_SCALE) */
static int initial_ssthresh = 100;
static int smooth_part = 20;

module_param(fast_convergence, int, 0644);
MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
module_param(max_increment, int, 0644);
MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
module_param(low_window, int, 0644);
MODULE_PARM_DESC(low_window, "lower bound on congestion window (for TCP friendliness)");
module_param(beta, int, 0644);
MODULE_PARM_DESC(beta, "beta for multiplicative increase");
module_param(initial_ssthresh, int, 0644);
MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
module_param(smooth_part, int, 0644);
MODULE_PARM_DESC(smooth_part, "log(B/(B*Smin))/log(B/(B-1))+B, # of RTT from Wmax-B to Wmax");


/* BIC TCP Parameters */
struct bictcp {
	u32	cnt;		/* increase cwnd by 1 after ACKs */
	u32 	last_max_cwnd;	/* last maximum snd_cwnd */
	u32	loss_cwnd;	/* congestion window at last loss */
	u32	last_cwnd;	/* the last snd_cwnd */
	u32	last_time;	/* time when updated last_cwnd */
	u32	epoch_start;	/* beginning of an epoch */
#define ACK_RATIO_SHIFT	4
	u32	delayed_ack;	/* estimate the ratio of Packets/ACKs << 4 */
};

static inline void bictcp_reset(struct bictcp *ca)
{
	ca->cnt = 0;
	ca->last_max_cwnd = 0;
	ca->loss_cwnd = 0;
	ca->last_cwnd = 0;
	ca->last_time = 0;
	ca->epoch_start = 0;
	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
}

static void bictcp_init(struct sock *sk)
{
	bictcp_reset(inet_csk_ca(sk));
	if (initial_ssthresh)
		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
}

/*
 * Compute congestion window to use.
 */
static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
{
	if (ca->last_cwnd == cwnd &&
	    (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
		return;

	ca->last_cwnd = cwnd;
	ca->last_time = tcp_time_stamp;

	if (ca->epoch_start == 0) /* record the beginning of an epoch */
		ca->epoch_start = tcp_time_stamp;

	/* start off normal */
	if (cwnd <= low_window) {
		ca->cnt = cwnd;
		return;
	}

	/* binary increase */
	if (cwnd < ca->last_max_cwnd) {
		__u32 	dist = (ca->last_max_cwnd - cwnd)
			/ BICTCP_B;

		if (dist > max_increment)
			/* linear increase */
			ca->cnt = cwnd / max_increment;
		else if (dist <= 1U)
			/* binary search increase */
			ca->cnt = (cwnd * smooth_part) / BICTCP_B;
		else
			/* binary search increase */
			ca->cnt = cwnd / dist;
	} else {
		/* slow start AMD linear increase */
		if (cwnd < ca->last_max_cwnd + BICTCP_B)
			/* slow start */
			ca->cnt = (cwnd * smooth_part) / BICTCP_B;
		else if (cwnd < ca->last_max_cwnd + max_increment*(BICTCP_B-1))
			/* slow start */
			ca->cnt = (cwnd * (BICTCP_B-1))
				/ (cwnd - ca->last_max_cwnd);
		else
			/* linear increase */
			ca->cnt = cwnd / max_increment;
	}

	/* if in slow start or link utilization is very low */
	if (ca->loss_cwnd == 0) {
		if (ca->cnt > 20) /* increase cwnd 5% per RTT */
			ca->cnt = 20;
	}

	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	if (ca->cnt == 0)			/* cannot be zero */
		ca->cnt = 1;
}

static void bictcp_cong_avoid(struct sock *sk, u32 ack,
			      u32 seq_rtt, u32 in_flight, int data_acked)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = 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 {
		bictcp_update(ca, tp->snd_cwnd);

		/* In dangerous area, increase slowly.
		 * In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
		 */
		if (tp->snd_cwnd_cnt >= ca->cnt) {
			if (tp->snd_cwnd < tp->snd_cwnd_clamp)
				tp->snd_cwnd++;
			tp->snd_cwnd_cnt = 0;
		} else
			tp->snd_cwnd_cnt++;
	}

}

/*
 *	behave like Reno until low_window is reached,
 *	then increase congestion window slowly
 */
static u32 bictcp_recalc_ssthresh(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	struct bictcp *ca = inet_csk_ca(sk);

	ca->epoch_start = 0;	/* end of epoch */

	/* Wmax and fast convergence */
	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
		ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
			/ (2 * BICTCP_BETA_SCALE);
	else
		ca->last_max_cwnd = tp->snd_cwnd;

	ca->loss_cwnd = tp->snd_cwnd;


	if (tp->snd_cwnd <= low_window)
		return max(tp->snd_cwnd >> 1U, 2U);
	else
		return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
}

static u32 bictcp_undo_cwnd(struct sock *sk)
{
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct bictcp *ca = inet_csk_ca(sk);
	return max(tp->snd_cwnd, ca->last_max_cwnd);
}

static void bictcp_state(struct sock *sk, u8 new_state)
{
	if (new_state == TCP_CA_Loss)
		bictcp_reset(inet_csk_ca(sk));
}

/* Track delayed acknowledgment ratio using sliding window
 * ratio = (15*ratio + sample) / 16
 */
static void bictcp_acked(struct sock *sk, u32 cnt)
{
	const struct inet_connection_sock *icsk = inet_csk(sk);

	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
		struct bictcp *ca = inet_csk_ca(sk);
		cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
		ca->delayed_ack += cnt;
	}
}


static struct tcp_congestion_ops bictcp = {
	.init		= bictcp_init,
	.ssthresh	= bictcp_recalc_ssthresh,
	.cong_avoid	= bictcp_cong_avoid,
	.set_state	= bictcp_state,
	.undo_cwnd	= bictcp_undo_cwnd,
	.pkts_acked     = bictcp_acked,
	.owner		= THIS_MODULE,
	.name		= "bic",
};

static int __init bictcp_register(void)
{
	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
	return tcp_register_congestion_control(&bictcp);
}

static void __exit bictcp_unregister(void)
{
	tcp_unregister_congestion_control(&bictcp);
}

module_init(bictcp_register);
module_exit(bictcp_unregister);

MODULE_AUTHOR("Stephen Hemminger");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("BIC TCP");
Commit	Line	Data
83803034 SH	1	/*
	2	* Binary Increase Congestion control for TCP
	3	*
	4	* This is from the implementation of BICTCP in
	5	* Lison-Xu, Kahaled Harfoush, and Injong Rhee.
	6	* "Binary Increase Congestion Control for Fast, Long Distance
	7	* Networks" in InfoComm 2004
	8	* Available from:
	9	* http://www.csc.ncsu.edu/faculty/rhee/export/bitcp.pdf
	10	*
	11	* Unless BIC is enabled and congestion window is large
	12	* this behaves the same as the original Reno.
	13	*/
	14
83803034 SH	15	#include <linux/mm.h>
	16	#include <linux/module.h>
	17	#include <net/tcp.h>
	18
	19
	20	#define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
	21	* max_cwnd = snd_cwnd * beta
	22	*/
	23	#define BICTCP_B 4 /*
	24	* In binary search,
	25	* go to point (max+min)/N
	26	*/
	27
	28	static int fast_convergence = 1;
450b5b18	29	static int max_increment = 16;
83803034 SH	30	static int low_window = 14;
83803034 SH	31	static int beta = 819; /* = 819/1024 (BICTCP_BETA_SCALE) */
83803034 SH	32	static int initial_ssthresh = 100;
	33	static int smooth_part = 20;
	34
	35	module_param(fast_convergence, int, 0644);
	36	MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
	37	module_param(max_increment, int, 0644);
	38	MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
	39	module_param(low_window, int, 0644);
	40	MODULE_PARM_DESC(low_window, "lower bound on congestion window (for TCP friendliness)");
	41	module_param(beta, int, 0644);
	42	MODULE_PARM_DESC(beta, "beta for multiplicative increase");
83803034 SH	43	module_param(initial_ssthresh, int, 0644);
	44	MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
	45	module_param(smooth_part, int, 0644);
	46	MODULE_PARM_DESC(smooth_part, "log(B/(B*Smin))/log(B/(B-1))+B, # of RTT from Wmax-B to Wmax");
	47
	48
	49	/* BIC TCP Parameters */
	50	struct bictcp {
	51	u32 cnt; /* increase cwnd by 1 after ACKs */
	52	u32 last_max_cwnd; /* last maximum snd_cwnd */
	53	u32 loss_cwnd; /* congestion window at last loss */
	54	u32 last_cwnd; /* the last snd_cwnd */
	55	u32 last_time; /* time when updated last_cwnd */
83803034 SH	56	u32 epoch_start; /* beginning of an epoch */
	57	#define ACK_RATIO_SHIFT 4
	58	u32 delayed_ack; /* estimate the ratio of Packets/ACKs << 4 */
	59	};
	60
	61	static inline void bictcp_reset(struct bictcp *ca)
	62	{
	63	ca->cnt = 0;
	64	ca->last_max_cwnd = 0;
	65	ca->loss_cwnd = 0;
	66	ca->last_cwnd = 0;
	67	ca->last_time = 0;
83803034 SH	68	ca->epoch_start = 0;
	69	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	70	}
	71
6687e988	72	static void bictcp_init(struct sock *sk)
83803034	73	{
6687e988	74	bictcp_reset(inet_csk_ca(sk));
83803034	75	if (initial_ssthresh)
6687e988	76	tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
83803034 SH	77	}
	78
	79	/*
	80	* Compute congestion window to use.
	81	*/
	82	static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
	83	{
	84	if (ca->last_cwnd == cwnd &&
	85	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	86	return;
	87
	88	ca->last_cwnd = cwnd;
	89	ca->last_time = tcp_time_stamp;
	90
	91	if (ca->epoch_start == 0) /* record the beginning of an epoch */
	92	ca->epoch_start = tcp_time_stamp;
	93
	94	/* start off normal */
	95	if (cwnd <= low_window) {
	96	ca->cnt = cwnd;
	97	return;
	98	}
	99
	100	/* binary increase */
	101	if (cwnd < ca->last_max_cwnd) {
	102	__u32 dist = (ca->last_max_cwnd - cwnd)
	103	/ BICTCP_B;
	104
	105	if (dist > max_increment)
	106	/* linear increase */
	107	ca->cnt = cwnd / max_increment;
	108	else if (dist <= 1U)
	109	/* binary search increase */
	110	ca->cnt = (cwnd * smooth_part) / BICTCP_B;
	111	else
	112	/* binary search increase */
	113	ca->cnt = cwnd / dist;
	114	} else {
	115	/* slow start AMD linear increase */
	116	if (cwnd < ca->last_max_cwnd + BICTCP_B)
	117	/* slow start */
	118	ca->cnt = (cwnd * smooth_part) / BICTCP_B;
	119	else if (cwnd < ca->last_max_cwnd + max_increment*(BICTCP_B-1))
	120	/* slow start */
	121	ca->cnt = (cwnd * (BICTCP_B-1))
42a39450	122	/ (cwnd - ca->last_max_cwnd);
83803034 SH	123	else
	124	/* linear increase */
	125	ca->cnt = cwnd / max_increment;
	126	}
	127
	128	/* if in slow start or link utilization is very low */
018da8f4	129	if (ca->loss_cwnd == 0) {
83803034 SH	130	if (ca->cnt > 20) /* increase cwnd 5% per RTT */
	131	ca->cnt = 20;
	132	}
	133
	134	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	135	if (ca->cnt == 0) /* cannot be zero */
	136	ca->cnt = 1;
	137	}
	138
6687e988	139	static void bictcp_cong_avoid(struct sock *sk, u32 ack,
83803034 SH	140	u32 seq_rtt, u32 in_flight, int data_acked)
83803034 SH	141	{
6687e988 ACM	142	struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	143	struct bictcp *ca = inet_csk_ca(sk);
83803034	144
f4805ede	145	if (!tcp_is_cwnd_limited(sk, in_flight))
83803034 SH	146	return;
83803034 SH	147
7faffa1c SH	148	if (tp->snd_cwnd <= tp->snd_ssthresh)
	149	tcp_slow_start(tp);
	150	else {
83803034 SH	151	bictcp_update(ca, tp->snd_cwnd);
83803034 SH	152
7faffa1c	153	/* In dangerous area, increase slowly.
83803034 SH	154	* In theory this is tp->snd_cwnd += 1 / tp->snd_cwnd
	155	*/
	156	if (tp->snd_cwnd_cnt >= ca->cnt) {
	157	if (tp->snd_cwnd < tp->snd_cwnd_clamp)
	158	tp->snd_cwnd++;
	159	tp->snd_cwnd_cnt = 0;
	160	} else
	161	tp->snd_cwnd_cnt++;
	162	}
	163
	164	}
	165
	166	/*
	167	* behave like Reno until low_window is reached,
	168	* then increase congestion window slowly
	169	*/
6687e988	170	static u32 bictcp_recalc_ssthresh(struct sock *sk)
83803034	171	{
6687e988 ACM	172	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	173	struct bictcp *ca = inet_csk_ca(sk);
83803034 SH	174
	175	ca->epoch_start = 0; /* end of epoch */
	176
83803034 SH	177	/* Wmax and fast convergence */
	178	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
	179	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
	180	/ (2 * BICTCP_BETA_SCALE);
	181	else
	182	ca->last_max_cwnd = tp->snd_cwnd;
	183
	184	ca->loss_cwnd = tp->snd_cwnd;
	185
	186
	187	if (tp->snd_cwnd <= low_window)
	188	return max(tp->snd_cwnd >> 1U, 2U);
	189	else
	190	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
	191	}
	192
6687e988	193	static u32 bictcp_undo_cwnd(struct sock *sk)
83803034	194	{
6687e988 ACM	195	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	196	const struct bictcp *ca = inet_csk_ca(sk);
83803034 SH	197	return max(tp->snd_cwnd, ca->last_max_cwnd);
	198	}
	199
6687e988	200	static void bictcp_state(struct sock *sk, u8 new_state)
83803034 SH	201	{
83803034 SH	202	if (new_state == TCP_CA_Loss)
6687e988	203	bictcp_reset(inet_csk_ca(sk));
83803034 SH	204	}
83803034 SH	205
05d05450	206	/* Track delayed acknowledgment ratio using sliding window
83803034 SH	207	* ratio = (15*ratio + sample) / 16
83803034 SH	208	*/
6687e988	209	static void bictcp_acked(struct sock *sk, u32 cnt)
83803034	210	{
6687e988 ACM	211	const struct inet_connection_sock *icsk = inet_csk(sk);
6687e988 ACM	212
05d05450	213	if (cnt > 0 && icsk->icsk_ca_state == TCP_CA_Open) {
6687e988	214	struct bictcp *ca = inet_csk_ca(sk);
83803034 SH	215	cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
	216	ca->delayed_ack += cnt;
	217	}
	218	}
	219
	220
	221	static struct tcp_congestion_ops bictcp = {
	222	.init = bictcp_init,
	223	.ssthresh = bictcp_recalc_ssthresh,
	224	.cong_avoid = bictcp_cong_avoid,
	225	.set_state = bictcp_state,
	226	.undo_cwnd = bictcp_undo_cwnd,
83803034 SH	227	.pkts_acked = bictcp_acked,
	228	.owner = THIS_MODULE,
	229	.name = "bic",
	230	};
	231
	232	static int __init bictcp_register(void)
	233	{
65e3d726	234	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
83803034 SH	235	return tcp_register_congestion_control(&bictcp);
	236	}
	237
	238	static void __exit bictcp_unregister(void)
	239	{
	240	tcp_unregister_congestion_control(&bictcp);
	241	}
	242
	243	module_init(bictcp_register);
	244	module_exit(bictcp_unregister);
	245
	246	MODULE_AUTHOR("Stephen Hemminger");
	247	MODULE_LICENSE("GPL");
	248	MODULE_DESCRIPTION("BIC TCP");