blob: 05c3b6f0e8e1cfa4b6f0530c826df9f5402dfe7d [file] [log] [blame]
Angelo P. Castellani5ef81472007-02-22 00:23:05 -08001/*
2 *
3 * YeAH TCP
4 *
5 * For further details look at:
6 * http://wil.cs.caltech.edu/pfldnet2007/paper/YeAH_TCP.pdf
7 *
8 */
Stephen Hemminger77522372007-04-23 22:28:23 -07009#include <linux/mm.h>
10#include <linux/module.h>
11#include <linux/skbuff.h>
12#include <linux/inet_diag.h>
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080013
Stephen Hemminger77522372007-04-23 22:28:23 -070014#include <net/tcp.h>
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080015
Stephen Hemminger77522372007-04-23 22:28:23 -070016#include "tcp_vegas.h"
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080017
18#define TCP_YEAH_ALPHA 80 //lin number of packets queued at the bottleneck
19#define TCP_YEAH_GAMMA 1 //lin fraction of queue to be removed per rtt
20#define TCP_YEAH_DELTA 3 //log minimum fraction of cwnd to be removed on loss
21#define TCP_YEAH_EPSILON 1 //log maximum fraction to be removed on early decongestion
22#define TCP_YEAH_PHY 8 //lin maximum delta from base
Lucas De Marchi25985ed2011-03-30 22:57:33 -030023#define TCP_YEAH_RHO 16 //lin minimum number of consecutive rtt to consider competition on loss
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080024#define TCP_YEAH_ZETA 50 //lin minimum number of state switchs to reset reno_count
25
26#define TCP_SCALABLE_AI_CNT 100U
27
28/* YeAH variables */
29struct yeah {
Stephen Hemminger77522372007-04-23 22:28:23 -070030 struct vegas vegas; /* must be first */
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080031
32 /* YeAH */
33 u32 lastQ;
34 u32 doing_reno_now;
35
36 u32 reno_count;
37 u32 fast_count;
38
39 u32 pkts_acked;
40};
41
42static void tcp_yeah_init(struct sock *sk)
43{
44 struct tcp_sock *tp = tcp_sk(sk);
45 struct yeah *yeah = inet_csk_ca(sk);
46
47 tcp_vegas_init(sk);
48
49 yeah->doing_reno_now = 0;
50 yeah->lastQ = 0;
51
52 yeah->reno_count = 2;
53
54 /* Ensure the MD arithmetic works. This is somewhat pedantic,
55 * since I don't think we will see a cwnd this large. :) */
56 tp->snd_cwnd_clamp = min_t(u32, tp->snd_cwnd_clamp, 0xffffffff/128);
57
58}
59
60
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -070061static void tcp_yeah_pkts_acked(struct sock *sk, u32 pkts_acked, s32 rtt_us)
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080062{
63 const struct inet_connection_sock *icsk = inet_csk(sk);
64 struct yeah *yeah = inet_csk_ca(sk);
65
66 if (icsk->icsk_ca_state == TCP_CA_Open)
67 yeah->pkts_acked = pkts_acked;
Stephen Hemminger164891a2007-04-23 22:26:16 -070068
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -070069 tcp_vegas_pkts_acked(sk, pkts_acked, rtt_us);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080070}
71
Ilpo Järvinenc3a05c62007-12-02 00:47:59 +020072static void tcp_yeah_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080073{
74 struct tcp_sock *tp = tcp_sk(sk);
75 struct yeah *yeah = inet_csk_ca(sk);
76
77 if (!tcp_is_cwnd_limited(sk, in_flight))
78 return;
79
Stephen Hemminger77522372007-04-23 22:28:23 -070080 if (tp->snd_cwnd <= tp->snd_ssthresh)
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080081 tcp_slow_start(tp);
Stephen Hemminger77522372007-04-23 22:28:23 -070082
83 else if (!yeah->doing_reno_now) {
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080084 /* Scalable */
85
Jianjun Kong5a5f3a82008-11-03 00:24:34 -080086 tp->snd_cwnd_cnt += yeah->pkts_acked;
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080087 if (tp->snd_cwnd_cnt > min(tp->snd_cwnd, TCP_SCALABLE_AI_CNT)){
88 if (tp->snd_cwnd < tp->snd_cwnd_clamp)
89 tp->snd_cwnd++;
90 tp->snd_cwnd_cnt = 0;
91 }
92
93 yeah->pkts_acked = 1;
94
95 } else {
96 /* Reno */
Ilpo Järvinen758ce5c2009-02-28 04:44:37 +000097 tcp_cong_avoid_ai(tp, tp->snd_cwnd);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -080098 }
99
Stephen Hemminger77522372007-04-23 22:28:23 -0700100 /* The key players are v_vegas.beg_snd_una and v_beg_snd_nxt.
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800101 *
102 * These are so named because they represent the approximate values
103 * of snd_una and snd_nxt at the beginning of the current RTT. More
104 * precisely, they represent the amount of data sent during the RTT.
105 * At the end of the RTT, when we receive an ACK for v_beg_snd_nxt,
Stephen Hemminger77522372007-04-23 22:28:23 -0700106 * we will calculate that (v_beg_snd_nxt - v_vegas.beg_snd_una) outstanding
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800107 * bytes of data have been ACKed during the course of the RTT, giving
108 * an "actual" rate of:
109 *
Stephen Hemminger77522372007-04-23 22:28:23 -0700110 * (v_beg_snd_nxt - v_vegas.beg_snd_una) / (rtt duration)
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800111 *
Stephen Hemminger77522372007-04-23 22:28:23 -0700112 * Unfortunately, v_vegas.beg_snd_una is not exactly equal to snd_una,
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800113 * because delayed ACKs can cover more than one segment, so they
114 * don't line up yeahly with the boundaries of RTTs.
115 *
116 * Another unfortunate fact of life is that delayed ACKs delay the
117 * advance of the left edge of our send window, so that the number
118 * of bytes we send in an RTT is often less than our cwnd will allow.
119 * So we keep track of our cwnd separately, in v_beg_snd_cwnd.
120 */
121
Stephen Hemminger77522372007-04-23 22:28:23 -0700122 if (after(ack, yeah->vegas.beg_snd_nxt)) {
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800123
124 /* We do the Vegas calculations only if we got enough RTT
125 * samples that we can be reasonably sure that we got
126 * at least one RTT sample that wasn't from a delayed ACK.
127 * If we only had 2 samples total,
128 * then that means we're getting only 1 ACK per RTT, which
129 * means they're almost certainly delayed ACKs.
130 * If we have 3 samples, we should be OK.
131 */
132
Stephen Hemminger77522372007-04-23 22:28:23 -0700133 if (yeah->vegas.cntRTT > 2) {
Stephen Hemminger43e68392007-03-06 20:21:20 -0800134 u32 rtt, queue;
135 u64 bw;
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800136
137 /* We have enough RTT samples, so, using the Vegas
138 * algorithm, we determine if we should increase or
139 * decrease cwnd, and by how much.
140 */
141
142 /* Pluck out the RTT we are using for the Vegas
143 * calculations. This is the min RTT seen during the
144 * last RTT. Taking the min filters out the effects
145 * of delayed ACKs, at the cost of noticing congestion
146 * a bit later.
147 */
Stephen Hemminger77522372007-04-23 22:28:23 -0700148 rtt = yeah->vegas.minRTT;
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800149
Stephen Hemminger43e68392007-03-06 20:21:20 -0800150 /* Compute excess number of packets above bandwidth
151 * Avoid doing full 64 bit divide.
152 */
153 bw = tp->snd_cwnd;
Stephen Hemminger77522372007-04-23 22:28:23 -0700154 bw *= rtt - yeah->vegas.baseRTT;
Stephen Hemminger43e68392007-03-06 20:21:20 -0800155 do_div(bw, rtt);
156 queue = bw;
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800157
Stephen Hemminger43e68392007-03-06 20:21:20 -0800158 if (queue > TCP_YEAH_ALPHA ||
Stephen Hemminger77522372007-04-23 22:28:23 -0700159 rtt - yeah->vegas.baseRTT > (yeah->vegas.baseRTT / TCP_YEAH_PHY)) {
Joe Perches9d4fb272009-11-23 10:41:23 -0800160 if (queue > TCP_YEAH_ALPHA &&
161 tp->snd_cwnd > yeah->reno_count) {
Stephen Hemminger43e68392007-03-06 20:21:20 -0800162 u32 reduction = min(queue / TCP_YEAH_GAMMA ,
163 tp->snd_cwnd >> TCP_YEAH_EPSILON);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800164
165 tp->snd_cwnd -= reduction;
166
Stephen Hemminger43e68392007-03-06 20:21:20 -0800167 tp->snd_cwnd = max(tp->snd_cwnd,
168 yeah->reno_count);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800169
170 tp->snd_ssthresh = tp->snd_cwnd;
Stephen Hemminger43e68392007-03-06 20:21:20 -0800171 }
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800172
173 if (yeah->reno_count <= 2)
Stephen Hemminger43e68392007-03-06 20:21:20 -0800174 yeah->reno_count = max(tp->snd_cwnd>>1, 2U);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800175 else
176 yeah->reno_count++;
177
Stephen Hemminger43e68392007-03-06 20:21:20 -0800178 yeah->doing_reno_now = min(yeah->doing_reno_now + 1,
179 0xffffffU);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800180 } else {
181 yeah->fast_count++;
182
183 if (yeah->fast_count > TCP_YEAH_ZETA) {
184 yeah->reno_count = 2;
185 yeah->fast_count = 0;
186 }
187
188 yeah->doing_reno_now = 0;
189 }
190
191 yeah->lastQ = queue;
192
193 }
194
195 /* Save the extent of the current window so we can use this
196 * at the end of the next RTT.
197 */
Stephen Hemminger77522372007-04-23 22:28:23 -0700198 yeah->vegas.beg_snd_una = yeah->vegas.beg_snd_nxt;
199 yeah->vegas.beg_snd_nxt = tp->snd_nxt;
200 yeah->vegas.beg_snd_cwnd = tp->snd_cwnd;
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800201
202 /* Wipe the slate clean for the next RTT. */
Stephen Hemminger77522372007-04-23 22:28:23 -0700203 yeah->vegas.cntRTT = 0;
204 yeah->vegas.minRTT = 0x7fffffff;
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800205 }
206}
207
208static u32 tcp_yeah_ssthresh(struct sock *sk) {
209 const struct tcp_sock *tp = tcp_sk(sk);
210 struct yeah *yeah = inet_csk_ca(sk);
211 u32 reduction;
212
213 if (yeah->doing_reno_now < TCP_YEAH_RHO) {
214 reduction = yeah->lastQ;
215
216 reduction = min( reduction, max(tp->snd_cwnd>>1, 2U) );
217
218 reduction = max( reduction, tp->snd_cwnd >> TCP_YEAH_DELTA);
219 } else
Jianjun Kong5a5f3a82008-11-03 00:24:34 -0800220 reduction = max(tp->snd_cwnd>>1, 2U);
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800221
222 yeah->fast_count = 0;
223 yeah->reno_count = max(yeah->reno_count>>1, 2U);
224
225 return tp->snd_cwnd - reduction;
226}
227
Stephen Hemmingera252beb2011-03-10 00:40:17 -0800228static struct tcp_congestion_ops tcp_yeah __read_mostly = {
Stephen Hemminger164891a2007-04-23 22:26:16 -0700229 .flags = TCP_CONG_RTT_STAMP,
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800230 .init = tcp_yeah_init,
231 .ssthresh = tcp_yeah_ssthresh,
232 .cong_avoid = tcp_yeah_cong_avoid,
233 .min_cwnd = tcp_reno_min_cwnd,
Angelo P. Castellani5ef81472007-02-22 00:23:05 -0800234 .set_state = tcp_vegas_state,
235 .cwnd_event = tcp_vegas_cwnd_event,
236 .get_info = tcp_vegas_get_info,
237 .pkts_acked = tcp_yeah_pkts_acked,
238
239 .owner = THIS_MODULE,
240 .name = "yeah",
241};
242
243static int __init tcp_yeah_register(void)
244{
245 BUG_ON(sizeof(struct yeah) > ICSK_CA_PRIV_SIZE);
246 tcp_register_congestion_control(&tcp_yeah);
247 return 0;
248}
249
250static void __exit tcp_yeah_unregister(void)
251{
252 tcp_unregister_congestion_control(&tcp_yeah);
253}
254
255module_init(tcp_yeah_register);
256module_exit(tcp_yeah_unregister);
257
258MODULE_AUTHOR("Angelo P. Castellani");
259MODULE_LICENSE("GPL");
260MODULE_DESCRIPTION("YeAH TCP");