blob: 4e1b61032a9cb4d7bcbae5ce4980cd1362f56353 [file] [log] [blame]
Stephen Hemminger87270762005-06-23 12:24:09 -07001/*
Luca De Ciccob7d7a9e2006-06-11 23:01:39 -07002 * TCP Westwood+: end-to-end bandwidth estimation for TCP
Stephen Hemminger87270762005-06-23 12:24:09 -07003 *
Luca De Ciccob7d7a9e2006-06-11 23:01:39 -07004 * Angelo Dell'Aera: author of the first version of TCP Westwood+ in Linux 2.4
5 *
6 * Support at http://c3lab.poliba.it/index.php/Westwood
7 * Main references in literature:
8 *
9 * - Mascolo S, Casetti, M. Gerla et al.
10 * "TCP Westwood: bandwidth estimation for TCP" Proc. ACM Mobicom 2001
11 *
12 * - A. Grieco, s. Mascolo
13 * "Performance evaluation of New Reno, Vegas, Westwood+ TCP" ACM Computer
14 * Comm. Review, 2004
15 *
16 * - A. Dell'Aera, L. Grieco, S. Mascolo.
17 * "Linux 2.4 Implementation of Westwood+ TCP with Rate-Halving :
18 * A Performance Evaluation Over the Internet" (ICC 2004), Paris, June 2004
19 *
20 * Westwood+ employs end-to-end bandwidth measurement to set cwnd and
21 * ssthresh after packet loss. The probing phase is as the original Reno.
Stephen Hemminger87270762005-06-23 12:24:09 -070022 */
23
Stephen Hemminger87270762005-06-23 12:24:09 -070024#include <linux/mm.h>
25#include <linux/module.h>
26#include <linux/skbuff.h>
Arnaldo Carvalho de Meloa8c21902005-08-12 12:56:38 -030027#include <linux/inet_diag.h>
Stephen Hemminger87270762005-06-23 12:24:09 -070028#include <net/tcp.h>
29
30/* TCP Westwood structure */
31struct westwood {
32 u32 bw_ns_est; /* first bandwidth estimation..not too smoothed 8) */
33 u32 bw_est; /* bandwidth estimate */
34 u32 rtt_win_sx; /* here starts a new evaluation... */
35 u32 bk;
36 u32 snd_una; /* used for evaluating the number of acked bytes */
37 u32 cumul_ack;
38 u32 accounted;
39 u32 rtt;
40 u32 rtt_min; /* minimum observed RTT */
Stephen Hemmingerf61e2902006-06-11 23:01:02 -070041 u8 first_ack; /* flag which infers that this is the first ack */
Luca De Ciccobc726a72006-06-11 23:02:19 -070042 u8 reset_rtt_min; /* Reset RTT min to next RTT sample*/
Stephen Hemminger87270762005-06-23 12:24:09 -070043};
44
45
46/* TCP Westwood functions and constants */
47#define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
48#define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
49
50/*
51 * @tcp_westwood_create
52 * This function initializes fields used in TCP Westwood+,
53 * it is called after the initial SYN, so the sequence numbers
54 * are correct but new passive connections we have no
55 * information about RTTmin at this time so we simply set it to
56 * TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
57 * since in this way we're sure it will be updated in a consistent
58 * way as soon as possible. It will reasonably happen within the first
59 * RTT period of the connection lifetime.
60 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030061static void tcp_westwood_init(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -070062{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030063 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -070064
65 w->bk = 0;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +090066 w->bw_ns_est = 0;
67 w->bw_est = 0;
68 w->accounted = 0;
69 w->cumul_ack = 0;
Luca De Ciccobc726a72006-06-11 23:02:19 -070070 w->reset_rtt_min = 1;
Stephen Hemminger87270762005-06-23 12:24:09 -070071 w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
72 w->rtt_win_sx = tcp_time_stamp;
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -030073 w->snd_una = tcp_sk(sk)->snd_una;
Stephen Hemmingerf61e2902006-06-11 23:01:02 -070074 w->first_ack = 1;
Stephen Hemminger87270762005-06-23 12:24:09 -070075}
76
77/*
78 * @westwood_do_filter
79 * Low-pass filter. Implemented using constant coefficients.
80 */
81static inline u32 westwood_do_filter(u32 a, u32 b)
82{
83 return (((7 * a) + b) >> 3);
84}
85
Luca De Ciccob3a92ea2006-06-11 23:01:59 -070086static void westwood_filter(struct westwood *w, u32 delta)
Stephen Hemminger87270762005-06-23 12:24:09 -070087{
Luca De Ciccob3a92ea2006-06-11 23:01:59 -070088 /* If the filter is empty fill it with the first sample of bandwidth */
89 if (w->bw_ns_est == 0 && w->bw_est == 0) {
90 w->bw_ns_est = w->bk / delta;
91 w->bw_est = w->bw_ns_est;
92 } else {
93 w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
94 w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
95 }
Stephen Hemminger87270762005-06-23 12:24:09 -070096}
97
98/*
99 * @westwood_pkts_acked
100 * Called after processing group of packets.
101 * but all westwood needs is the last sample of srtt.
102 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300103static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt)
Stephen Hemminger87270762005-06-23 12:24:09 -0700104{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300105 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700106 if (cnt > 0)
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300107 w->rtt = tcp_sk(sk)->srtt >> 3;
Stephen Hemminger87270762005-06-23 12:24:09 -0700108}
109
110/*
111 * @westwood_update_window
112 * It updates RTT evaluation window if it is the right moment to do
113 * it. If so it calls filter for evaluating bandwidth.
114 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300115static void westwood_update_window(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700116{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300117 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700118 s32 delta = tcp_time_stamp - w->rtt_win_sx;
119
Luca De Ciccob7d7a9e2006-06-11 23:01:39 -0700120 /* Initialize w->snd_una with the first acked sequence number in order
Stephen Hemmingerf61e2902006-06-11 23:01:02 -0700121 * to fix mismatch between tp->snd_una and w->snd_una for the first
122 * bandwidth sample
123 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900124 if (w->first_ack) {
Stephen Hemmingerf61e2902006-06-11 23:01:02 -0700125 w->snd_una = tcp_sk(sk)->snd_una;
126 w->first_ack = 0;
127 }
128
Stephen Hemminger87270762005-06-23 12:24:09 -0700129 /*
130 * See if a RTT-window has passed.
131 * Be careful since if RTT is less than
132 * 50ms we don't filter but we continue 'building the sample'.
133 * This minimum limit was chosen since an estimation on small
134 * time intervals is better to avoid...
135 * Obviously on a LAN we reasonably will always have
136 * right_bound = left_bound + WESTWOOD_RTT_MIN
137 */
138 if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
139 westwood_filter(w, delta);
140
141 w->bk = 0;
142 w->rtt_win_sx = tcp_time_stamp;
143 }
144}
145
Luca De Ciccobc726a72006-06-11 23:02:19 -0700146static inline void update_rtt_min(struct westwood *w)
147{
148 if (w->reset_rtt_min) {
149 w->rtt_min = w->rtt;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900150 w->reset_rtt_min = 0;
Luca De Ciccobc726a72006-06-11 23:02:19 -0700151 } else
152 w->rtt_min = min(w->rtt, w->rtt_min);
153}
154
155
Stephen Hemminger87270762005-06-23 12:24:09 -0700156/*
157 * @westwood_fast_bw
158 * It is called when we are in fast path. In particular it is called when
159 * header prediction is successful. In such case in fact update is
160 * straight forward and doesn't need any particular care.
161 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300162static inline void westwood_fast_bw(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700163{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300164 const struct tcp_sock *tp = tcp_sk(sk);
165 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700166
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300167 westwood_update_window(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700168
169 w->bk += tp->snd_una - w->snd_una;
170 w->snd_una = tp->snd_una;
Luca De Ciccobc726a72006-06-11 23:02:19 -0700171 update_rtt_min(w);
Stephen Hemminger87270762005-06-23 12:24:09 -0700172}
173
174/*
175 * @westwood_acked_count
176 * This function evaluates cumul_ack for evaluating bk in case of
177 * delayed or partial acks.
178 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300179static inline u32 westwood_acked_count(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700180{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300181 const struct tcp_sock *tp = tcp_sk(sk);
182 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700183
184 w->cumul_ack = tp->snd_una - w->snd_una;
185
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900186 /* If cumul_ack is 0 this is a dupack since it's not moving
187 * tp->snd_una.
188 */
189 if (!w->cumul_ack) {
Stephen Hemminger87270762005-06-23 12:24:09 -0700190 w->accounted += tp->mss_cache;
191 w->cumul_ack = tp->mss_cache;
192 }
193
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900194 if (w->cumul_ack > tp->mss_cache) {
Stephen Hemminger87270762005-06-23 12:24:09 -0700195 /* Partial or delayed ack */
196 if (w->accounted >= w->cumul_ack) {
197 w->accounted -= w->cumul_ack;
198 w->cumul_ack = tp->mss_cache;
199 } else {
200 w->cumul_ack -= w->accounted;
201 w->accounted = 0;
202 }
203 }
204
205 w->snd_una = tp->snd_una;
206
207 return w->cumul_ack;
208}
209
Stephen Hemminger87270762005-06-23 12:24:09 -0700210
211/*
212 * TCP Westwood
213 * Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
214 * in packets we use mss_cache). Rttmin is guaranteed to be >= 2
215 * so avoids ever returning 0.
216 */
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700217static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700218{
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700219 const struct tcp_sock *tp = tcp_sk(sk);
220 const struct westwood *w = inet_csk_ca(sk);
221 return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
Stephen Hemminger87270762005-06-23 12:24:09 -0700222}
223
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300224static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
Stephen Hemminger87270762005-06-23 12:24:09 -0700225{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300226 struct tcp_sock *tp = tcp_sk(sk);
227 struct westwood *w = inet_csk_ca(sk);
Luca De Ciccob7d7a9e2006-06-11 23:01:39 -0700228
Stephen Hemminger87270762005-06-23 12:24:09 -0700229 switch(event) {
230 case CA_EVENT_FAST_ACK:
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300231 westwood_fast_bw(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700232 break;
233
234 case CA_EVENT_COMPLETE_CWR:
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700235 tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700236 break;
237
238 case CA_EVENT_FRTO:
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700239 tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900240 /* Update RTT_min when next ack arrives */
Luca De Ciccobc726a72006-06-11 23:02:19 -0700241 w->reset_rtt_min = 1;
Stephen Hemminger87270762005-06-23 12:24:09 -0700242 break;
243
244 case CA_EVENT_SLOW_ACK:
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300245 westwood_update_window(sk);
246 w->bk += westwood_acked_count(sk);
Luca De Ciccobc726a72006-06-11 23:02:19 -0700247 update_rtt_min(w);
Stephen Hemminger87270762005-06-23 12:24:09 -0700248 break;
249
250 default:
251 /* don't care */
252 break;
253 }
254}
255
256
257/* Extract info for Tcp socket info provided via netlink. */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300258static void tcp_westwood_info(struct sock *sk, u32 ext,
Stephen Hemminger87270762005-06-23 12:24:09 -0700259 struct sk_buff *skb)
260{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300261 const struct westwood *ca = inet_csk_ca(sk);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300262 if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
Stephen Hemminger87270762005-06-23 12:24:09 -0700263 struct rtattr *rta;
264 struct tcpvegas_info *info;
265
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300266 rta = __RTA_PUT(skb, INET_DIAG_VEGASINFO, sizeof(*info));
Stephen Hemminger87270762005-06-23 12:24:09 -0700267 info = RTA_DATA(rta);
268 info->tcpv_enabled = 1;
269 info->tcpv_rttcnt = 0;
270 info->tcpv_rtt = jiffies_to_usecs(ca->rtt);
271 info->tcpv_minrtt = jiffies_to_usecs(ca->rtt_min);
272 rtattr_failure: ;
273 }
274}
275
276
277static struct tcp_congestion_ops tcp_westwood = {
278 .init = tcp_westwood_init,
279 .ssthresh = tcp_reno_ssthresh,
280 .cong_avoid = tcp_reno_cong_avoid,
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700281 .min_cwnd = tcp_westwood_bw_rttmin,
Stephen Hemminger87270762005-06-23 12:24:09 -0700282 .cwnd_event = tcp_westwood_event,
283 .get_info = tcp_westwood_info,
284 .pkts_acked = tcp_westwood_pkts_acked,
285
286 .owner = THIS_MODULE,
287 .name = "westwood"
288};
289
290static int __init tcp_westwood_register(void)
291{
Alexey Dobriyan74975d42006-08-25 17:10:33 -0700292 BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
Stephen Hemminger87270762005-06-23 12:24:09 -0700293 return tcp_register_congestion_control(&tcp_westwood);
294}
295
296static void __exit tcp_westwood_unregister(void)
297{
298 tcp_unregister_congestion_control(&tcp_westwood);
299}
300
301module_init(tcp_westwood_register);
302module_exit(tcp_westwood_unregister);
303
304MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
305MODULE_LICENSE("GPL");
306MODULE_DESCRIPTION("TCP Westwood+");