blob: 76a1e23259e1fa713bb447486e4e94848e419d41 [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{
Eric Dumazeta02cec22010-09-22 20:43:57 +000083 return ((7 * a) + b) >> 3;
Stephen Hemminger87270762005-06-23 12:24:09 -070084}
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 */
Stephen Hemminger30cfd0b2007-07-25 23:49:34 -0700103static void tcp_westwood_pkts_acked(struct sock *sk, u32 cnt, s32 rtt)
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 Hemminger30cfd0b2007-07-25 23:49:34 -0700106
107 if (rtt > 0)
108 w->rtt = usecs_to_jiffies(rtt);
Stephen Hemminger87270762005-06-23 12:24:09 -0700109}
110
111/*
112 * @westwood_update_window
113 * It updates RTT evaluation window if it is the right moment to do
114 * it. If so it calls filter for evaluating bandwidth.
115 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300116static void westwood_update_window(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700117{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300118 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700119 s32 delta = tcp_time_stamp - w->rtt_win_sx;
120
Luca De Ciccob7d7a9e2006-06-11 23:01:39 -0700121 /* Initialize w->snd_una with the first acked sequence number in order
Stephen Hemmingerf61e2902006-06-11 23:01:02 -0700122 * to fix mismatch between tp->snd_una and w->snd_una for the first
123 * bandwidth sample
124 */
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900125 if (w->first_ack) {
Stephen Hemmingerf61e2902006-06-11 23:01:02 -0700126 w->snd_una = tcp_sk(sk)->snd_una;
127 w->first_ack = 0;
128 }
129
Stephen Hemminger87270762005-06-23 12:24:09 -0700130 /*
131 * See if a RTT-window has passed.
132 * Be careful since if RTT is less than
133 * 50ms we don't filter but we continue 'building the sample'.
134 * This minimum limit was chosen since an estimation on small
135 * time intervals is better to avoid...
136 * Obviously on a LAN we reasonably will always have
137 * right_bound = left_bound + WESTWOOD_RTT_MIN
138 */
139 if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
140 westwood_filter(w, delta);
141
142 w->bk = 0;
143 w->rtt_win_sx = tcp_time_stamp;
144 }
145}
146
Luca De Ciccobc726a72006-06-11 23:02:19 -0700147static inline void update_rtt_min(struct westwood *w)
148{
149 if (w->reset_rtt_min) {
150 w->rtt_min = w->rtt;
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900151 w->reset_rtt_min = 0;
Luca De Ciccobc726a72006-06-11 23:02:19 -0700152 } else
153 w->rtt_min = min(w->rtt, w->rtt_min);
154}
155
156
Stephen Hemminger87270762005-06-23 12:24:09 -0700157/*
158 * @westwood_fast_bw
159 * It is called when we are in fast path. In particular it is called when
160 * header prediction is successful. In such case in fact update is
161 * straight forward and doesn't need any particular care.
162 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300163static inline void westwood_fast_bw(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700164{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300165 const struct tcp_sock *tp = tcp_sk(sk);
166 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700167
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300168 westwood_update_window(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700169
170 w->bk += tp->snd_una - w->snd_una;
171 w->snd_una = tp->snd_una;
Luca De Ciccobc726a72006-06-11 23:02:19 -0700172 update_rtt_min(w);
Stephen Hemminger87270762005-06-23 12:24:09 -0700173}
174
175/*
176 * @westwood_acked_count
177 * This function evaluates cumul_ack for evaluating bk in case of
178 * delayed or partial acks.
179 */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300180static inline u32 westwood_acked_count(struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700181{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300182 const struct tcp_sock *tp = tcp_sk(sk);
183 struct westwood *w = inet_csk_ca(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700184
185 w->cumul_ack = tp->snd_una - w->snd_una;
186
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900187 /* If cumul_ack is 0 this is a dupack since it's not moving
188 * tp->snd_una.
189 */
190 if (!w->cumul_ack) {
Stephen Hemminger87270762005-06-23 12:24:09 -0700191 w->accounted += tp->mss_cache;
192 w->cumul_ack = tp->mss_cache;
193 }
194
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900195 if (w->cumul_ack > tp->mss_cache) {
Stephen Hemminger87270762005-06-23 12:24:09 -0700196 /* Partial or delayed ack */
197 if (w->accounted >= w->cumul_ack) {
198 w->accounted -= w->cumul_ack;
199 w->cumul_ack = tp->mss_cache;
200 } else {
201 w->cumul_ack -= w->accounted;
202 w->accounted = 0;
203 }
204 }
205
206 w->snd_una = tp->snd_una;
207
208 return w->cumul_ack;
209}
210
Stephen Hemminger87270762005-06-23 12:24:09 -0700211
212/*
213 * TCP Westwood
214 * Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
215 * in packets we use mss_cache). Rttmin is guaranteed to be >= 2
216 * so avoids ever returning 0.
217 */
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700218static u32 tcp_westwood_bw_rttmin(const struct sock *sk)
Stephen Hemminger87270762005-06-23 12:24:09 -0700219{
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700220 const struct tcp_sock *tp = tcp_sk(sk);
221 const struct westwood *w = inet_csk_ca(sk);
222 return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
Stephen Hemminger87270762005-06-23 12:24:09 -0700223}
224
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300225static void tcp_westwood_event(struct sock *sk, enum tcp_ca_event event)
Stephen Hemminger87270762005-06-23 12:24:09 -0700226{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300227 struct tcp_sock *tp = tcp_sk(sk);
228 struct westwood *w = inet_csk_ca(sk);
Luca De Ciccob7d7a9e2006-06-11 23:01:39 -0700229
Stephen Hemminger2de979b2007-03-08 20:45:19 -0800230 switch (event) {
Stephen Hemminger87270762005-06-23 12:24:09 -0700231 case CA_EVENT_FAST_ACK:
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300232 westwood_fast_bw(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700233 break;
234
235 case CA_EVENT_COMPLETE_CWR:
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700236 tp->snd_cwnd = tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
Stephen Hemminger87270762005-06-23 12:24:09 -0700237 break;
238
Yuchung Cheng9b441902013-03-20 13:32:58 +0000239 case CA_EVENT_LOSS:
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700240 tp->snd_ssthresh = tcp_westwood_bw_rttmin(sk);
YOSHIFUJI Hideakie905a9e2007-02-09 23:24:47 +0900241 /* Update RTT_min when next ack arrives */
Luca De Ciccobc726a72006-06-11 23:02:19 -0700242 w->reset_rtt_min = 1;
Stephen Hemminger87270762005-06-23 12:24:09 -0700243 break;
244
245 case CA_EVENT_SLOW_ACK:
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300246 westwood_update_window(sk);
247 w->bk += westwood_acked_count(sk);
Luca De Ciccobc726a72006-06-11 23:02:19 -0700248 update_rtt_min(w);
Stephen Hemminger87270762005-06-23 12:24:09 -0700249 break;
250
251 default:
252 /* don't care */
253 break;
254 }
255}
256
257
258/* Extract info for Tcp socket info provided via netlink. */
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300259static void tcp_westwood_info(struct sock *sk, u32 ext,
Stephen Hemminger87270762005-06-23 12:24:09 -0700260 struct sk_buff *skb)
261{
Arnaldo Carvalho de Melo6687e982005-08-10 04:03:31 -0300262 const struct westwood *ca = inet_csk_ca(sk);
Arnaldo Carvalho de Melo73c1f4a2005-08-12 12:51:49 -0300263 if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
Thomas Graf26728102007-03-22 23:27:19 -0700264 struct tcpvegas_info info = {
265 .tcpv_enabled = 1,
266 .tcpv_rtt = jiffies_to_usecs(ca->rtt),
267 .tcpv_minrtt = jiffies_to_usecs(ca->rtt_min),
268 };
Stephen Hemminger87270762005-06-23 12:24:09 -0700269
Thomas Graf26728102007-03-22 23:27:19 -0700270 nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
Stephen Hemminger87270762005-06-23 12:24:09 -0700271 }
272}
273
274
Stephen Hemmingera252beb2011-03-10 00:40:17 -0800275static struct tcp_congestion_ops tcp_westwood __read_mostly = {
Stephen Hemminger87270762005-06-23 12:24:09 -0700276 .init = tcp_westwood_init,
277 .ssthresh = tcp_reno_ssthresh,
278 .cong_avoid = tcp_reno_cong_avoid,
Stephen Hemminger72dc5b92006-06-05 17:30:08 -0700279 .min_cwnd = tcp_westwood_bw_rttmin,
Stephen Hemminger87270762005-06-23 12:24:09 -0700280 .cwnd_event = tcp_westwood_event,
281 .get_info = tcp_westwood_info,
282 .pkts_acked = tcp_westwood_pkts_acked,
283
284 .owner = THIS_MODULE,
285 .name = "westwood"
286};
287
288static int __init tcp_westwood_register(void)
289{
Alexey Dobriyan74975d42006-08-25 17:10:33 -0700290 BUILD_BUG_ON(sizeof(struct westwood) > ICSK_CA_PRIV_SIZE);
Stephen Hemminger87270762005-06-23 12:24:09 -0700291 return tcp_register_congestion_control(&tcp_westwood);
292}
293
294static void __exit tcp_westwood_unregister(void)
295{
296 tcp_unregister_congestion_control(&tcp_westwood);
297}
298
299module_init(tcp_westwood_register);
300module_exit(tcp_westwood_unregister);
301
302MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
303MODULE_LICENSE("GPL");
304MODULE_DESCRIPTION("TCP Westwood+");