blob: f45e1c24244091a750f5f34ca0fe9e9038731ca5 [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;
Stephen Hemminger83803032005-06-23 12:23:25 -070066 ca->last_cwnd = 0;
67 ca->last_time = 0;
Stephen Hemminger83803032005-06-23 12:23:25 -070068 ca->epoch_start = 0;
69 ca->delayed_ack = 2 << ACK_RATIO_SHIFT;
70}
71
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030072static void bictcp_init(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -070073{
Neal Cardwellfc16dcd2012-01-18 17:47:58 +000074 struct bictcp *ca = inet_csk_ca(sk);
75
76 bictcp_reset(ca);
77 ca->loss_cwnd = 0;
78
Stephen Hemminger83803032005-06-23 12:23:25 -070079 if (initial_ssthresh)
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030080 tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
Stephen Hemminger83803032005-06-23 12:23:25 -070081}
82
83/*
84 * Compute congestion window to use.
85 */
86static inline void bictcp_update(struct bictcp *ca, u32 cwnd)
87{
88 if (ca->last_cwnd == cwnd &&
89 (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
90 return;
91
92 ca->last_cwnd = cwnd;
93 ca->last_time = tcp_time_stamp;
94
95 if (ca->epoch_start == 0) /* record the beginning of an epoch */
96 ca->epoch_start = tcp_time_stamp;
97
98 /* start off normal */
99 if (cwnd <= low_window) {
100 ca->cnt = cwnd;
101 return;
102 }
103
104 /* binary increase */
105 if (cwnd < ca->last_max_cwnd) {
106 __u32 dist = (ca->last_max_cwnd - cwnd)
107 / BICTCP_B;
108
109 if (dist > max_increment)
110 /* linear increase */
111 ca->cnt = cwnd / max_increment;
112 else if (dist <= 1U)
113 /* binary search increase */
114 ca->cnt = (cwnd * smooth_part) / BICTCP_B;
115 else
116 /* binary search increase */
117 ca->cnt = cwnd / dist;
118 } else {
119 /* slow start AMD linear increase */
120 if (cwnd < ca->last_max_cwnd + BICTCP_B)
121 /* slow start */
122 ca->cnt = (cwnd * smooth_part) / BICTCP_B;
123 else if (cwnd < ca->last_max_cwnd + max_increment*(BICTCP_B-1))
124 /* slow start */
125 ca->cnt = (cwnd * (BICTCP_B-1))
Stephen Hemminger42a39452005-10-05 12:09:31 -0700126 / (cwnd - ca->last_max_cwnd);
Stephen Hemminger83803032005-06-23 12:23:25 -0700127 else
128 /* linear increase */
129 ca->cnt = cwnd / max_increment;
130 }
131
132 /* if in slow start or link utilization is very low */
Neal Cardwellfc16dcd2012-01-18 17:47:58 +0000133 if (ca->last_max_cwnd == 0) {
Stephen Hemminger83803032005-06-23 12:23:25 -0700134 if (ca->cnt > 20) /* increase cwnd 5% per RTT */
135 ca->cnt = 20;
136 }
137
138 ca->cnt = (ca->cnt << ACK_RATIO_SHIFT) / ca->delayed_ack;
139 if (ca->cnt == 0) /* cannot be zero */
140 ca->cnt = 1;
141}
142
Ilpo Järvinenc3a05c62007-12-02 00:47:59 +0200143static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
Stephen Hemminger83803032005-06-23 12:23:25 -0700144{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300145 struct tcp_sock *tp = tcp_sk(sk);
146 struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700147
Stephen Hemmingerf4805ed2005-11-10 16:53:30 -0800148 if (!tcp_is_cwnd_limited(sk, in_flight))
Stephen Hemminger83803032005-06-23 12:23:25 -0700149 return;
150
Stephen Hemminger7faffa12005-11-10 17:07:24 -0800151 if (tp->snd_cwnd <= tp->snd_ssthresh)
152 tcp_slow_start(tp);
153 else {
Stephen Hemminger83803032005-06-23 12:23:25 -0700154 bictcp_update(ca, tp->snd_cwnd);
Ilpo Järvinen758ce5c2009-02-28 04:44:37 +0000155 tcp_cong_avoid_ai(tp, ca->cnt);
Stephen Hemminger83803032005-06-23 12:23:25 -0700156 }
157
158}
159
160/*
161 * behave like Reno until low_window is reached,
162 * then increase congestion window slowly
163 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300164static u32 bictcp_recalc_ssthresh(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -0700165{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300166 const struct tcp_sock *tp = tcp_sk(sk);
167 struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700168
169 ca->epoch_start = 0; /* end of epoch */
170
Stephen Hemminger83803032005-06-23 12:23:25 -0700171 /* Wmax and fast convergence */
172 if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
173 ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
174 / (2 * BICTCP_BETA_SCALE);
175 else
176 ca->last_max_cwnd = tp->snd_cwnd;
177
178 ca->loss_cwnd = tp->snd_cwnd;
179
180
181 if (tp->snd_cwnd <= low_window)
182 return max(tp->snd_cwnd >> 1U, 2U);
183 else
184 return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
185}
186
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300187static u32 bictcp_undo_cwnd(struct sock *sk)
Stephen Hemminger83803032005-06-23 12:23:25 -0700188{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300189 const struct tcp_sock *tp = tcp_sk(sk);
190 const struct bictcp *ca = inet_csk_ca(sk);
Neal Cardwellfc16dcd2012-01-18 17:47:58 +0000191 return max(tp->snd_cwnd, ca->loss_cwnd);
Stephen Hemminger83803032005-06-23 12:23:25 -0700192}
193
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300194static void bictcp_state(struct sock *sk, u8 new_state)
Stephen Hemminger83803032005-06-23 12:23:25 -0700195{
196 if (new_state == TCP_CA_Loss)
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300197 bictcp_reset(inet_csk_ca(sk));
Stephen Hemminger83803032005-06-23 12:23:25 -0700198}
199
Stephen Hemminger05d05452005-12-13 23:13:13 -0800200/* Track delayed acknowledgment ratio using sliding window
Stephen Hemminger83803032005-06-23 12:23:25 -0700201 * ratio = (15*ratio + sample) / 16
202 */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700203static void bictcp_acked(struct sock *sk, u32 cnt, s32 rtt)
Stephen Hemminger83803032005-06-23 12:23:25 -0700204{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300205 const struct inet_connection_sock *icsk = inet_csk(sk);
206
Ilpo Järvinen35e86942007-05-31 10:16:47 +0300207 if (icsk->icsk_ca_state == TCP_CA_Open) {
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300208 struct bictcp *ca = inet_csk_ca(sk);
Stephen Hemminger83803032005-06-23 12:23:25 -0700209 cnt -= ca->delayed_ack >> ACK_RATIO_SHIFT;
210 ca->delayed_ack += cnt;
211 }
212}
213
214
Stephen Hemmingera252beb2011-03-10 00:40:17 -0800215static struct tcp_congestion_ops bictcp __read_mostly = {
Stephen Hemminger83803032005-06-23 12:23:25 -0700216 .init = bictcp_init,
217 .ssthresh = bictcp_recalc_ssthresh,
218 .cong_avoid = bictcp_cong_avoid,
219 .set_state = bictcp_state,
220 .undo_cwnd = bictcp_undo_cwnd,
Stephen Hemminger83803032005-06-23 12:23:25 -0700221 .pkts_acked = bictcp_acked,
222 .owner = THIS_MODULE,
223 .name = "bic",
224};
225
226static int __init bictcp_register(void)
227{
Alexey Dobriyan65e3d722006-08-25 00:38:03 -0700228 BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
Stephen Hemminger83803032005-06-23 12:23:25 -0700229 return tcp_register_congestion_control(&bictcp);
230}
231
232static void __exit bictcp_unregister(void)
233{
234 tcp_unregister_congestion_control(&bictcp);
235}
236
237module_init(bictcp_register);
238module_exit(bictcp_unregister);
239
240MODULE_AUTHOR("Stephen Hemminger");
241MODULE_LICENSE("GPL");
242MODULE_DESCRIPTION("BIC TCP");