aboutsummaryrefslogtreecommitdiff
path: root/net/ipv4/tcp_yeah.c
blob: 028eb046ea40d510908040023b46b46469477505 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
/*
 *
 *   YeAH TCP
 *
 * For further details look at:
 *   https://web.archive.org/web/20080316215752/http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf
 *
 */
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/skbuff.h>
#include <linux/inet_diag.h>

#include <net/tcp.h>

#include "tcp_vegas.h"

#define TCP_YEAH_ALPHA       80 /* number of packets queued at the bottleneck */
#define TCP_YEAH_GAMMA        1 /* 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 /* maximum delta from base */
#define TCP_YEAH_RHO         16 /* minimum number of consecutive rtt to consider competition on loss */
#define TCP_YEAH_ZETA        50 /* minimum number of state switches to reset reno_count */

#define TCP_SCALABLE_AI_CNT	 100U

/* YeAH variables */
struct yeah {
	struct vegas vegas;	/* must be first */

	/* 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,
				const struct ack_sample *sample)
{
	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 = sample->pkts_acked;

	tcp_vegas_pkts_acked(sk, sample);
}

static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack, u32 acked)
{
	struct tcp_sock *tp = tcp_sk(sk);
	struct yeah *yeah = inet_csk_ca(sk);

	if (!tcp_is_cwnd_limited(sk))
		return;

	if (tp->snd_cwnd <= tp->snd_ssthresh)
		tcp_slow_start(tp, acked);

	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 */
		tcp_cong_avoid_ai(tp, tp->snd_cwnd, 1);
	}

	/* The key players are v_vegas.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_vegas.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_vegas.beg_snd_una) / (rtt duration)
	 *
	 * Unfortunately, v_vegas.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->vegas.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->vegas.cntRTT > 2) {
			u32 rtt, queue;
			u64 bw;

			/* 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->vegas.minRTT;

			/* Compute excess number of packets above bandwidth
			 * Avoid doing full 64 bit divide.
			 */
			bw = tp->snd_cwnd;
			bw *= rtt - yeah->vegas.baseRTT;
			do_div(bw, rtt);
			queue = bw;

			if (queue > TCP_YEAH_ALPHA ||
			    rtt - yeah->vegas.baseRTT > (yeah->vegas.baseRTT / TCP_YEAH_PHY)) {
				if (queue > TCP_YEAH_ALPHA &&
				    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(yeah->doing_reno_now + 1,
							   0xffffffU);
			} 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->vegas.beg_snd_una  = yeah->vegas.beg_snd_nxt;
		yeah->vegas.beg_snd_nxt  = tp->snd_nxt;
		yeah->vegas.beg_snd_cwnd = tp->snd_cwnd;

		/* Wipe the slate clean for the next RTT. */
		yeah->vegas.cntRTT = 0;
		yeah->vegas.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 max_t(int, tp->snd_cwnd - reduction, 2);
}

static struct tcp_congestion_ops tcp_yeah __read_mostly = {
	.init		= tcp_yeah_init,
	.ssthresh	= tcp_yeah_ssthresh,
	.cong_avoid	= tcp_yeah_cong_avoid,
	.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");