[net-next-2.6.git] / net / ipv4 / tcp_bic.c

/*
 * Binary Increase Congestion control for TCP
 * Home page:
 *      http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
 * 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://netsrv.csc.ncsu.edu/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;
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 in_flight)
{
	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);
		tcp_cong_avoid_ai(tp, ca->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, s32 rtt)
{
	const struct inet_connection_sock *icsk = inet_csk(sk);

	if (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	/*
83803034 SH	2	* Binary Increase Congestion control for TCP
0bc8c7bf SH	3	* Home page:
0bc8c7bf SH	4	* http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
83803034 SH	5	* This is from the implementation of BICTCP in
	6	* Lison-Xu, Kahaled Harfoush, and Injong Rhee.
	7	* "Binary Increase Congestion Control for Fast, Long Distance
	8	* Networks" in InfoComm 2004
	9	* Available from:
0bc8c7bf	10	* http://netsrv.csc.ncsu.edu/export/bitcp.pdf
83803034 SH	11	*
	12	* Unless BIC is enabled and congestion window is large
	13	* this behaves the same as the original Reno.
	14	*/
	15
83803034 SH	16	#include <linux/mm.h>
	17	#include <linux/module.h>
	18	#include <net/tcp.h>
	19
	20
	21	#define BICTCP_BETA_SCALE 1024 /* Scale factor beta calculation
	22	* max_cwnd = snd_cwnd * beta
	23	*/
	24	#define BICTCP_B 4 /*
	25	* In binary search,
	26	* go to point (max+min)/N
	27	*/
	28
	29	static int fast_convergence = 1;
450b5b18	30	static int max_increment = 16;
83803034 SH	31	static int low_window = 14;
83803034 SH	32	static int beta = 819; /* = 819/1024 (BICTCP_BETA_SCALE) */
66e1e3b2	33	static int initial_ssthresh;
83803034 SH	34	static int smooth_part = 20;
	35
	36	module_param(fast_convergence, int, 0644);
	37	MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
	38	module_param(max_increment, int, 0644);
	39	MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
	40	module_param(low_window, int, 0644);
	41	MODULE_PARM_DESC(low_window, "lower bound on congestion window (for TCP friendliness)");
	42	module_param(beta, int, 0644);
	43	MODULE_PARM_DESC(beta, "beta for multiplicative increase");
83803034 SH	44	module_param(initial_ssthresh, int, 0644);
	45	MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
	46	module_param(smooth_part, int, 0644);
	47	MODULE_PARM_DESC(smooth_part, "log(B/(B*Smin))/log(B/(B-1))+B, # of RTT from Wmax-B to Wmax");
	48
	49
	50	/* BIC TCP Parameters */
	51	struct bictcp {
	52	u32 cnt; /* increase cwnd by 1 after ACKs */
	53	u32 last_max_cwnd; /* last maximum snd_cwnd */
	54	u32 loss_cwnd; /* congestion window at last loss */
	55	u32 last_cwnd; /* the last snd_cwnd */
	56	u32 last_time; /* time when updated last_cwnd */
83803034 SH	57	u32 epoch_start; /* beginning of an epoch */
	58	#define ACK_RATIO_SHIFT 4
	59	u32 delayed_ack; /* estimate the ratio of Packets/ACKs << 4 */
	60	};
	61
	62	static inline void bictcp_reset(struct bictcp *ca)
	63	{
	64	ca->cnt = 0;
	65	ca->last_max_cwnd = 0;
	66	ca->loss_cwnd = 0;
	67	ca->last_cwnd = 0;
	68	ca->last_time = 0;
83803034 SH	69	ca->epoch_start = 0;
	70	ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
	71	}
	72
6687e988	73	static void bictcp_init(struct sock *sk)
83803034	74	{
6687e988	75	bictcp_reset(inet_csk_ca(sk));
83803034	76	if (initial_ssthresh)
6687e988	77	tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
83803034 SH	78	}
	79
	80	/*
	81	* Compute congestion window to use.
	82	*/
	83	static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
	84	{
	85	if (ca->last_cwnd == cwnd &&
	86	(s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
	87	return;
	88
	89	ca->last_cwnd = cwnd;
	90	ca->last_time = tcp_time_stamp;
	91
	92	if (ca->epoch_start == 0) /* record the beginning of an epoch */
	93	ca->epoch_start = tcp_time_stamp;
	94
	95	/* start off normal */
	96	if (cwnd <= low_window) {
	97	ca->cnt = cwnd;
	98	return;
	99	}
	100
	101	/* binary increase */
	102	if (cwnd < ca->last_max_cwnd) {
	103	__u32 dist = (ca->last_max_cwnd - cwnd)
	104	/ BICTCP_B;
	105
	106	if (dist > max_increment)
	107	/* linear increase */
	108	ca->cnt = cwnd / max_increment;
	109	else if (dist <= 1U)
	110	/* binary search increase */
	111	ca->cnt = (cwnd * smooth_part) / BICTCP_B;
	112	else
	113	/* binary search increase */
	114	ca->cnt = cwnd / dist;
	115	} else {
	116	/* slow start AMD linear increase */
	117	if (cwnd < ca->last_max_cwnd + BICTCP_B)
	118	/* slow start */
	119	ca->cnt = (cwnd * smooth_part) / BICTCP_B;
	120	else if (cwnd < ca->last_max_cwnd + max_increment*(BICTCP_B-1))
	121	/* slow start */
	122	ca->cnt = (cwnd * (BICTCP_B-1))
42a39450	123	/ (cwnd - ca->last_max_cwnd);
83803034 SH	124	else
	125	/* linear increase */
	126	ca->cnt = cwnd / max_increment;
	127	}
	128
	129	/* if in slow start or link utilization is very low */
018da8f4	130	if (ca->loss_cwnd == 0) {
83803034 SH	131	if (ca->cnt > 20) /* increase cwnd 5% per RTT */
	132	ca->cnt = 20;
	133	}
	134
	135	ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
	136	if (ca->cnt == 0) /* cannot be zero */
	137	ca->cnt = 1;
	138	}
	139
c3a05c60	140	static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
83803034	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	151	bictcp_update(ca, tp->snd_cwnd);
758ce5c8	152	tcp_cong_avoid_ai(tp, ca->cnt);
83803034 SH	153	}
	154
	155	}
	156
	157	/*
	158	* behave like Reno until low_window is reached,
	159	* then increase congestion window slowly
	160	*/
6687e988	161	static u32 bictcp_recalc_ssthresh(struct sock *sk)
83803034	162	{
6687e988 ACM	163	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	164	struct bictcp *ca = inet_csk_ca(sk);
83803034 SH	165
	166	ca->epoch_start = 0; /* end of epoch */
	167
83803034 SH	168	/* Wmax and fast convergence */
	169	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
	170	ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
	171	/ (2 * BICTCP_BETA_SCALE);
	172	else
	173	ca->last_max_cwnd = tp->snd_cwnd;
	174
	175	ca->loss_cwnd = tp->snd_cwnd;
	176
	177
	178	if (tp->snd_cwnd <= low_window)
	179	return max(tp->snd_cwnd >> 1U, 2U);
	180	else
	181	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
	182	}
	183
6687e988	184	static u32 bictcp_undo_cwnd(struct sock *sk)
83803034	185	{
6687e988 ACM	186	const struct tcp_sock *tp = tcp_sk(sk);
6687e988 ACM	187	const struct bictcp *ca = inet_csk_ca(sk);
83803034 SH	188	return max(tp->snd_cwnd, ca->last_max_cwnd);
	189	}
	190
6687e988	191	static void bictcp_state(struct sock *sk, u8 new_state)
83803034 SH	192	{
83803034 SH	193	if (new_state == TCP_CA_Loss)
6687e988	194	bictcp_reset(inet_csk_ca(sk));
83803034 SH	195	}
83803034 SH	196
05d05450	197	/* Track delayed acknowledgment ratio using sliding window
83803034 SH	198	* ratio = (15*ratio + sample) / 16
83803034 SH	199	*/
30cfd0ba	200	static void bictcp_acked(struct sock *sk, u32 cnt, s32 rtt)
83803034	201	{
6687e988 ACM	202	const struct inet_connection_sock *icsk = inet_csk(sk);
6687e988 ACM	203
35e86941	204	if (icsk->icsk_ca_state == TCP_CA_Open) {
6687e988	205	struct bictcp *ca = inet_csk_ca(sk);
83803034 SH	206	cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
	207	ca->delayed_ack += cnt;
	208	}
	209	}
	210
	211
	212	static struct tcp_congestion_ops bictcp = {
	213	.init = bictcp_init,
	214	.ssthresh = bictcp_recalc_ssthresh,
	215	.cong_avoid = bictcp_cong_avoid,
	216	.set_state = bictcp_state,
	217	.undo_cwnd = bictcp_undo_cwnd,
83803034 SH	218	.pkts_acked = bictcp_acked,
	219	.owner = THIS_MODULE,
	220	.name = "bic",
	221	};
	222
	223	static int __init bictcp_register(void)
	224	{
65e3d726	225	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
83803034 SH	226	return tcp_register_congestion_control(&bictcp);
	227	}
	228
	229	static void __exit bictcp_unregister(void)
	230	{
	231	tcp_unregister_congestion_control(&bictcp);
	232	}
	233
	234	module_init(bictcp_register);
	235	module_exit(bictcp_unregister);
	236
	237	MODULE_AUTHOR("Stephen Hemminger");
	238	MODULE_LICENSE("GPL");
	239	MODULE_DESCRIPTION("BIC TCP");