tcp_westwood.c 8.0 KB
Newer Older
1
/*
L
Luca De Cicco 已提交
2
 * TCP Westwood+: end-to-end bandwidth estimation for TCP
3
 *
L
Luca De Cicco 已提交
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
 *      Angelo Dell'Aera: author of the first version of TCP Westwood+ in Linux 2.4
 *
 * Support at http://c3lab.poliba.it/index.php/Westwood
 * Main references in literature:
 *
 * - Mascolo S, Casetti, M. Gerla et al.
 *   "TCP Westwood: bandwidth estimation for TCP" Proc. ACM Mobicom 2001
 *
 * - A. Grieco, s. Mascolo
 *   "Performance evaluation of New Reno, Vegas, Westwood+ TCP" ACM Computer
 *     Comm. Review, 2004
 *
 * - A. Dell'Aera, L. Grieco, S. Mascolo.
 *   "Linux 2.4 Implementation of Westwood+ TCP with Rate-Halving :
 *    A Performance Evaluation Over the Internet" (ICC 2004), Paris, June 2004
 *
 * Westwood+ employs end-to-end bandwidth measurement to set cwnd and
 * ssthresh after packet loss. The probing phase is as the original Reno.
22 23 24 25 26
 */

#include <linux/mm.h>
#include <linux/module.h>
#include <linux/skbuff.h>
27
#include <linux/inet_diag.h>
28 29 30 31 32 33 34 35 36 37 38 39 40
#include <net/tcp.h>

/* TCP Westwood structure */
struct westwood {
	u32    bw_ns_est;        /* first bandwidth estimation..not too smoothed 8) */
	u32    bw_est;           /* bandwidth estimate */
	u32    rtt_win_sx;       /* here starts a new evaluation... */
	u32    bk;
	u32    snd_una;          /* used for evaluating the number of acked bytes */
	u32    cumul_ack;
	u32    accounted;
	u32    rtt;
	u32    rtt_min;          /* minimum observed RTT */
41
	u8     first_ack;        /* flag which infers that this is the first ack */
42
	u8     reset_rtt_min;    /* Reset RTT min to next RTT sample*/
43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
};


/* TCP Westwood functions and constants */
#define TCP_WESTWOOD_RTT_MIN   (HZ/20)	/* 50ms */
#define TCP_WESTWOOD_INIT_RTT  (20*HZ)	/* maybe too conservative?! */

/*
 * @tcp_westwood_create
 * This function initializes fields used in TCP Westwood+,
 * it is called after the initial SYN, so the sequence numbers
 * are correct but new passive connections we have no
 * information about RTTmin at this time so we simply set it to
 * TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
 * since in this way we're sure it will be updated in a consistent
 * way as soon as possible. It will reasonably happen within the first
 * RTT period of the connection lifetime.
 */
61
static void tcp_westwood_init(struct sock *sk)
62
{
63
	struct westwood *w = inet_csk_ca(sk);
64 65

	w->bk = 0;
66 67 68 69
	w->bw_ns_est = 0;
	w->bw_est = 0;
	w->accounted = 0;
	w->cumul_ack = 0;
70
	w->reset_rtt_min = 1;
71 72
	w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
	w->rtt_win_sx = tcp_time_stamp;
73
	w->snd_una = tcp_sk(sk)->snd_una;
74
	w->first_ack = 1;
75 76 77 78 79 80 81 82 83 84 85
}

/*
 * @westwood_do_filter
 * Low-pass filter. Implemented using constant coefficients.
 */
static inline u32 westwood_do_filter(u32 a, u32 b)
{
	return (((7 * a) + b) >> 3);
}

86
static void westwood_filter(struct westwood *w, u32 delta)
87
{
88 89 90 91 92 93 94 95
	/* If the filter is empty fill it with the first sample of bandwidth  */
	if (w->bw_ns_est == 0 && w->bw_est == 0) {
		w->bw_ns_est = w->bk / delta;
		w->bw_est = w->bw_ns_est;
	} else {
		w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
		w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
	}
96 97 98 99 100 101 102
}

/*
 * @westwood_pkts_acked
 * Called after processing group of packets.
 * but all westwood needs is the last sample of srtt.
 */
103
static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt, ktime_t last)
104
{
105
	struct westwood *w = inet_csk_ca(sk);
106
	if (cnt > 0)
107
		w->rtt = tcp_sk(sk)->srtt >> 3;
108 109 110 111 112 113 114
}

/*
 * @westwood_update_window
 * It updates RTT evaluation window if it is the right moment to do
 * it. If so it calls filter for evaluating bandwidth.
 */
115
static void westwood_update_window(struct sock *sk)
116
{
117
	struct westwood *w = inet_csk_ca(sk);
118 119
	s32 delta = tcp_time_stamp - w->rtt_win_sx;

L
Luca De Cicco 已提交
120
	/* Initialize w->snd_una with the first acked sequence number in order
121 122 123
	 * to fix mismatch between tp->snd_una and w->snd_una for the first
	 * bandwidth sample
	 */
124
	if (w->first_ack) {
125 126 127 128
		w->snd_una = tcp_sk(sk)->snd_una;
		w->first_ack = 0;
	}

129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145
	/*
	 * See if a RTT-window has passed.
	 * Be careful since if RTT is less than
	 * 50ms we don't filter but we continue 'building the sample'.
	 * This minimum limit was chosen since an estimation on small
	 * time intervals is better to avoid...
	 * Obviously on a LAN we reasonably will always have
	 * right_bound = left_bound + WESTWOOD_RTT_MIN
	 */
	if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
		westwood_filter(w, delta);

		w->bk = 0;
		w->rtt_win_sx = tcp_time_stamp;
	}
}

146 147 148 149
static inline void update_rtt_min(struct westwood *w)
{
	if (w->reset_rtt_min) {
		w->rtt_min = w->rtt;
150
		w->reset_rtt_min = 0;
151 152 153 154 155
	} else
		w->rtt_min = min(w->rtt, w->rtt_min);
}


156 157 158 159 160 161
/*
 * @westwood_fast_bw
 * It is called when we are in fast path. In particular it is called when
 * header prediction is successful. In such case in fact update is
 * straight forward and doesn't need any particular care.
 */
162
static inline void westwood_fast_bw(struct sock *sk)
163
{
164 165
	const struct tcp_sock *tp = tcp_sk(sk);
	struct westwood *w = inet_csk_ca(sk);
166

167
	westwood_update_window(sk);
168 169 170

	w->bk += tp->snd_una - w->snd_una;
	w->snd_una = tp->snd_una;
171
	update_rtt_min(w);
172 173 174 175 176 177 178
}

/*
 * @westwood_acked_count
 * This function evaluates cumul_ack for evaluating bk in case of
 * delayed or partial acks.
 */
179
static inline u32 westwood_acked_count(struct sock *sk)
180
{
181 182
	const struct tcp_sock *tp = tcp_sk(sk);
	struct westwood *w = inet_csk_ca(sk);
183 184 185

	w->cumul_ack = tp->snd_una - w->snd_una;

186 187 188 189
	/* If cumul_ack is 0 this is a dupack since it's not moving
	 * tp->snd_una.
	 */
	if (!w->cumul_ack) {
190 191 192 193
		w->accounted += tp->mss_cache;
		w->cumul_ack = tp->mss_cache;
	}

194
	if (w->cumul_ack > tp->mss_cache) {
195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216
		/* Partial or delayed ack */
		if (w->accounted >= w->cumul_ack) {
			w->accounted -= w->cumul_ack;
			w->cumul_ack = tp->mss_cache;
		} else {
			w->cumul_ack -= w->accounted;
			w->accounted = 0;
		}
	}

	w->snd_una = tp->snd_una;

	return w->cumul_ack;
}


/*
 * TCP Westwood
 * Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
 * in packets we use mss_cache). Rttmin is guaranteed to be >= 2
 * so avoids ever returning 0.
 */
217
static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
218
{
219 220 221
	const struct tcp_sock *tp = tcp_sk(sk);
	const struct westwood *w = inet_csk_ca(sk);
	return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
222 223
}

224
static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
225
{
226 227
	struct tcp_sock *tp = tcp_sk(sk);
	struct westwood *w = inet_csk_ca(sk);
L
Luca De Cicco 已提交
228

S
Stephen Hemminger 已提交
229
	switch (event) {
230
	case CA_EVENT_FAST_ACK:
231
		westwood_fast_bw(sk);
232 233 234
		break;

	case CA_EVENT_COMPLETE_CWR:
235
		tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
236 237 238
		break;

	case CA_EVENT_FRTO:
239
		tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
240
		/* Update RTT_min when next ack arrives */
241
		w->reset_rtt_min = 1;
242 243 244
		break;

	case CA_EVENT_SLOW_ACK:
245 246
		westwood_update_window(sk);
		w->bk += westwood_acked_count(sk);
247
		update_rtt_min(w);
248 249 250 251 252 253 254 255 256 257
		break;

	default:
		/* don't care */
		break;
	}
}


/* Extract info for Tcp socket info provided via netlink. */
258
static void tcp_westwood_info(struct sock *sk, u32 ext,
259 260
			      struct sk_buff *skb)
{
261
	const struct westwood *ca = inet_csk_ca(sk);
262
	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
263 264 265 266 267 268 269
		struct tcpvegas_info info = {
			.tcpv_enabled = 1,
			.tcpv_rtt = jiffies_to_usecs(ca->rtt),
			.tcpv_minrtt = jiffies_to_usecs(ca->rtt_min),
		};

		nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
270 271 272 273 274 275 276 277
	}
}


static struct tcp_congestion_ops tcp_westwood = {
	.init		= tcp_westwood_init,
	.ssthresh	= tcp_reno_ssthresh,
	.cong_avoid	= tcp_reno_cong_avoid,
278
	.min_cwnd	= tcp_westwood_bw_rttmin,
279 280 281 282 283 284 285 286 287 288
	.cwnd_event	= tcp_westwood_event,
	.get_info	= tcp_westwood_info,
	.pkts_acked	= tcp_westwood_pkts_acked,

	.owner		= THIS_MODULE,
	.name		= "westwood"
};

static int __init tcp_westwood_register(void)
{
289
	BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
290 291 292 293 294 295 296 297 298 299 300 301 302 303
	return tcp_register_congestion_control(&tcp_westwood);
}

static void __exit tcp_westwood_unregister(void)
{
	tcp_unregister_congestion_control(&tcp_westwood);
}

module_init(tcp_westwood_register);
module_exit(tcp_westwood_unregister);

MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
MODULE_LICENSE("GPL");
MODULE_DESCRIPTION("TCP Westwood+");