blob: 3b53fd1af23f04bb2a9023b56204ca45051fa81f [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
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
29static int fast_convergence = 1;
Stephen Hemminger450b5b12005-11-01 15:26:45 -080030static int max_increment = 16;
Stephen Hemminger83803032005-06-23 12:23:25 -070031static int low_window = 14;
32static int beta = 819; /* = 819/1024 (BICTCP_BETA_SCALE) */
David S. Miller66e1e3b2007-06-13 01:03:53 -070033static int initial_ssthresh;
Stephen Hemminger83803032005-06-23 12:23:25 -070034static int smooth_part = 20;
35
36module_param(fast_convergence, int, 0644);
37MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
38module_param(max_increment, int, 0644);
39MODULE_PARM_DESC(max_increment, "Limit on increment allowed during binary search");
40module_param(low_window, int, 0644);
41MODULE_PARM_DESC(low_window, "lower bound on congestion window (for TCP friendliness)");
42module_param(beta, int, 0644);
43MODULE_PARM_DESC(beta, "beta for multiplicative increase");
Stephen Hemminger83803032005-06-23 12:23:25 -070044module_param(initial_ssthresh, int, 0644);
45MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
46module_param(smooth_part, int, 0644);
47MODULE_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 */
51struct 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 */
Stephen Hemminger83803032005-06-23 12:23:25 -070057 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
62static 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;
Stephen Hemminger83803032005-06-23 12:23:25 -070069 ca->epoch_start = 0;
70 ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
71}
72
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030073static void bictcp_init(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -070074{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030075 bictcp_reset(inet_csk_ca(sk));
Stephen Hemminger83803032005-06-23 12:23:25 -070076 if (initial_ssthresh)
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030077 tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
Stephen Hemminger83803032005-06-23 12:23:25 -070078}
79
80/*
81 * Compute congestion window to use.
82 */
83static 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))
Stephen Hemminger42a39452005-10-05 12:09:31 -0700123 / (cwnd - ca->last_max_cwnd);
Stephen Hemminger83803032005-06-23 12:23:25 -0700124 else
125 /* linear increase */
126 ca->cnt = cwnd / max_increment;
127 }
128
129 /* if in slow start or link utilization is very low */
Stephen Hemminger018da8f2005-12-13 23:13:00 -0800130 if (ca->loss_cwnd == 0) {
Stephen Hemminger83803032005-06-23 12:23:25 -0700131 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
Ilpo Järvinenc3a05c62007-12-02 00:47:59 +0200140static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
Stephen Hemminger83803032005-06-23 12:23:25 -0700141{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300142 struct tcp_sock *tp = tcp_sk(sk);
143 struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700144
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800145 if (!tcp_is_cwnd_limited(sk, in_flight))
Stephen Hemminger83803032005-06-23 12:23:25 -0700146 return;
147
Stephen Hemminger7faffa12005-11-10 17:07:24 -0800148 if (tp->snd_cwnd <= tp->snd_ssthresh)
149 tcp_slow_start(tp);
150 else {
Stephen Hemminger83803032005-06-23 12:23:25 -0700151 bictcp_update(ca, tp->snd_cwnd);
Ilpo Järvinen758ce5c2009-02-28 04:44:37 +0000152 tcp_cong_avoid_ai(tp, ca->cnt);
Stephen Hemminger83803032005-06-23 12:23:25 -0700153 }
154
155}
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
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
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300184static u32 bictcp_undo_cwnd(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -0700185{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300186 const struct tcp_sock *tp = tcp_sk(sk);
187 const struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700188 return max(tp->snd_cwnd, ca->last_max_cwnd);
189}
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 */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700200static void bictcp_acked(struct sock *sk, u32 cnt, s32 rtt)
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 Hemminger83803032005-06-23 12:23:25 -0700206 cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
207 ca->delayed_ack += cnt;
208 }
209}
210
211
212static 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,
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");