blob: 36087bca9f489646c2ca5aae3111449a956dd33b [file] [log] [blame]
Stephen Hemminger83803032005-06-23 12:23:25 -07001/*
2 * Binary Increase Congestion control for TCP
Sangtae Ha0bc8c7b2008-02-28 22:14:32 -08003 * Home page:
4 * http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
Stephen Hemminger83803032005-06-23 12:23:25 -07005 * 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:
Sangtae Ha0bc8c7b2008-02-28 22:14:32 -080010 * http://netsrv.csc.ncsu.edu/export/bitcp.pdf
Stephen Hemminger83803032005-06-23 12:23:25 -070011 *
12 * Unless BIC is enabled and congestion window is large
13 * this behaves the same as the original Reno.
14 */
15
Stephen Hemminger83803032005-06-23 12:23:25 -070016#include <linux/mm.h>
17#include <linux/module.h>
18#include <net/tcp.h>
19
Stephen Hemminger83803032005-06-23 12:23:25 -070020#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
28static int fast_convergence = 1;
Stephen Hemminger450b5b12005-11-01 15:26:45 -080029static int max_increment = 16;
Stephen Hemminger83803032005-06-23 12:23:25 -070030static int low_window = 14;
31static int beta = 819; /* = 819/1024 (BICTCP_BETA_SCALE) */
David S. Miller66e1e3b2007-06-13 01:03:53 -070032static int initial_ssthresh;
Stephen Hemminger83803032005-06-23 12:23:25 -070033static int smooth_part = 20;
34
35module_param(fast_convergence, int, 0644);
36MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
37module_param(max_increment, int, 0644);
38MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
39module_param(low_window, int, 0644);
40MODULE_PARM_DESC(low_window, "lower bound on congestion window (for TCP friendliness)");
41module_param(beta, int, 0644);
42MODULE_PARM_DESC(beta, "beta for multiplicative increase");
Stephen Hemminger83803032005-06-23 12:23:25 -070043module_param(initial_ssthresh, int, 0644);
44MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
45module_param(smooth_part, int, 0644);
46MODULE_PARM_DESC(smooth_part, "log(B/(B*Smin))/log(B/(B-1))+B, # of RTT from Wmax-B to Wmax");
47
Stephen Hemminger83803032005-06-23 12:23:25 -070048/* BIC TCP Parameters */
49struct bictcp {
50 u32 cnt; /* increase cwnd by 1 after ACKs */
stephen hemminger688d1942014-08-29 23:32:05 -070051 u32 last_max_cwnd; /* last maximum snd_cwnd */
Stephen Hemminger83803032005-06-23 12:23:25 -070052 u32 loss_cwnd; /* congestion window at last loss */
53 u32 last_cwnd; /* the last snd_cwnd */
54 u32 last_time; /* time when updated last_cwnd */
Stephen Hemminger83803032005-06-23 12:23:25 -070055 u32 epoch_start; /* beginning of an epoch */
56#define ACK_RATIO_SHIFT 4
57 u32 delayed_ack; /* estimate the ratio of Packets/ACKs << 4 */
58};
59
60static inline void bictcp_reset(struct bictcp *ca)
61{
62 ca->cnt = 0;
63 ca->last_max_cwnd = 0;
Stephen Hemminger83803032005-06-23 12:23:25 -070064 ca->last_cwnd = 0;
65 ca->last_time = 0;
Stephen Hemminger83803032005-06-23 12:23:25 -070066 ca->epoch_start = 0;
67 ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
68}
69
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030070static void bictcp_init(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -070071{
Neal Cardwellfc16dcd2012-01-18 17:47:58 +000072 struct bictcp *ca = inet_csk_ca(sk);
73
74 bictcp_reset(ca);
75 ca->loss_cwnd = 0;
76
Stephen Hemminger83803032005-06-23 12:23:25 -070077 if (initial_ssthresh)
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030078 tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
Stephen Hemminger83803032005-06-23 12:23:25 -070079}
80
81/*
82 * Compute congestion window to use.
83 */
84static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
85{
86 if (ca->last_cwnd == cwnd &&
87 (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
88 return;
89
90 ca->last_cwnd = cwnd;
91 ca->last_time = tcp_time_stamp;
92
93 if (ca->epoch_start == 0) /* record the beginning of an epoch */
94 ca->epoch_start = tcp_time_stamp;
95
96 /* start off normal */
97 if (cwnd <= low_window) {
98 ca->cnt = cwnd;
99 return;
100 }
101
102 /* binary increase */
103 if (cwnd < ca->last_max_cwnd) {
stephen hemminger688d1942014-08-29 23:32:05 -0700104 __u32 dist = (ca->last_max_cwnd - cwnd)
Stephen Hemminger83803032005-06-23 12:23:25 -0700105 / BICTCP_B;
106
107 if (dist > max_increment)
108 /* linear increase */
109 ca->cnt = cwnd / max_increment;
110 else if (dist <= 1U)
111 /* binary search increase */
112 ca->cnt = (cwnd * smooth_part) / BICTCP_B;
113 else
114 /* binary search increase */
115 ca->cnt = cwnd / dist;
116 } else {
117 /* slow start AMD linear increase */
118 if (cwnd < ca->last_max_cwnd + BICTCP_B)
119 /* slow start */
120 ca->cnt = (cwnd * smooth_part) / BICTCP_B;
121 else if (cwnd < ca->last_max_cwnd + max_increment*(BICTCP_B-1))
122 /* slow start */
123 ca->cnt = (cwnd * (BICTCP_B-1))
Stephen Hemminger42a39452005-10-05 12:09:31 -0700124 / (cwnd - ca->last_max_cwnd);
Stephen Hemminger83803032005-06-23 12:23:25 -0700125 else
126 /* linear increase */
127 ca->cnt = cwnd / max_increment;
128 }
129
130 /* if in slow start or link utilization is very low */
Neal Cardwellfc16dcd2012-01-18 17:47:58 +0000131 if (ca->last_max_cwnd == 0) {
Stephen Hemminger83803032005-06-23 12:23:25 -0700132 if (ca->cnt > 20) /* increase cwnd 5% per RTT */
133 ca->cnt = 20;
134 }
135
136 ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
137 if (ca->cnt == 0) /* cannot be zero */
138 ca->cnt = 1;
139}
140
Eric Dumazet24901552014-05-02 21:18:05 -0700141static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
Stephen Hemminger83803032005-06-23 12:23:25 -0700142{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300143 struct tcp_sock *tp = tcp_sk(sk);
144 struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700145
Eric Dumazet24901552014-05-02 21:18:05 -0700146 if (!tcp_is_cwnd_limited(sk))
Stephen Hemminger83803032005-06-23 12:23:25 -0700147 return;
148
Yuchung Cheng071d5082015-07-09 13:16:29 -0700149 if (tcp_in_slow_start(tp))
Yuchung Cheng9f9843a72013-10-31 11:07:31 -0700150 tcp_slow_start(tp, acked);
Stephen Hemminger7faffa12005-11-10 17:07:24 -0800151 else {
Stephen Hemminger83803032005-06-23 12:23:25 -0700152 bictcp_update(ca, tp->snd_cwnd);
Neal Cardwelle73ebb02015-01-28 20:01:35 -0500153 tcp_cong_avoid_ai(tp, ca->cnt, 1);
Stephen Hemminger83803032005-06-23 12:23:25 -0700154 }
Stephen Hemminger83803032005-06-23 12:23:25 -0700155}
156
157/*
158 * behave like Reno until low_window is reached,
159 * then increase congestion window slowly
160 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300161static u32 bictcp_recalc_ssthresh(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -0700162{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300163 const struct tcp_sock *tp = tcp_sk(sk);
164 struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700165
166 ca->epoch_start = 0; /* end of epoch */
167
Stephen Hemminger83803032005-06-23 12:23:25 -0700168 /* 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
Stephen Hemminger83803032005-06-23 12:23:25 -0700177 if (tp->snd_cwnd <= low_window)
178 return max(tp->snd_cwnd >> 1U, 2U);
179 else
180 return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
181}
182
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300183static u32 bictcp_undo_cwnd(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -0700184{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300185 const struct tcp_sock *tp = tcp_sk(sk);
186 const struct bictcp *ca = inet_csk_ca(sk);
stephen hemminger688d1942014-08-29 23:32:05 -0700187
Neal Cardwellfc16dcd2012-01-18 17:47:58 +0000188 return max(tp->snd_cwnd, ca->loss_cwnd);
Stephen Hemminger83803032005-06-23 12:23:25 -0700189}
190
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300191static void bictcp_state(struct sock *sk, u8 new_state)
Stephen Hemminger83803032005-06-23 12:23:25 -0700192{
193 if (new_state == TCP_CA_Loss)
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300194 bictcp_reset(inet_csk_ca(sk));
Stephen Hemminger83803032005-06-23 12:23:25 -0700195}
196
Stephen Hemminger05d05452005-12-13 23:13:13 -0800197/* Track delayed acknowledgment ratio using sliding window
Stephen Hemminger83803032005-06-23 12:23:25 -0700198 * ratio = (15*ratio + sample) / 16
199 */
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700200static void bictcp_acked(struct sock *sk, const struct ack_sample *sample)
Stephen Hemminger83803032005-06-23 12:23:25 -0700201{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300202 const struct inet_connection_sock *icsk = inet_csk(sk);
203
Ilpo Järvinen35e86942007-05-31 10:16:47 +0300204 if (icsk->icsk_ca_state == TCP_CA_Open) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300205 struct bictcp *ca = inet_csk_ca(sk);
stephen hemminger688d1942014-08-29 23:32:05 -0700206
Lawrence Brakmo756ee172016-05-11 10:02:13 -0700207 ca->delayed_ack += sample->pkts_acked -
208 (ca->delayed_ack >> ACK_RATIO_SHIFT);
Stephen Hemminger83803032005-06-23 12:23:25 -0700209 }
210}
211
Stephen Hemmingera252beb2011-03-10 00:40:17 -0800212static struct tcp_congestion_ops bictcp __read_mostly = {
Stephen Hemminger83803032005-06-23 12:23:25 -0700213 .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,
Stephen Hemminger83803032005-06-23 12:23:25 -0700218 .pkts_acked = bictcp_acked,
219 .owner = THIS_MODULE,
220 .name = "bic",
221};
222
223static int __init bictcp_register(void)
224{
Alexey Dobriyan65e3d722006-08-25 00:38:03 -0700225 BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
Stephen Hemminger83803032005-06-23 12:23:25 -0700226 return tcp_register_congestion_control(&bictcp);
227}
228
229static void __exit bictcp_unregister(void)
230{
231 tcp_unregister_congestion_control(&bictcp);
232}
233
234module_init(bictcp_register);
235module_exit(bictcp_unregister);
236
237MODULE_AUTHOR("Stephen Hemminger");
238MODULE_LICENSE("GPL");
239MODULE_DESCRIPTION("BIC TCP");