aboutsummaryrefslogtreecommitdiffstats
path: root/net
diff options
context:
space:
mode:
authorStephen Hemminger <shemminger@osdl.org>2005-06-23 15:24:09 -0400
committerDavid S. Miller <davem@davemloft.net>2005-06-23 15:24:09 -0400
commit8727076289ec55298a05cabddf02b374d13c1624 (patch)
tree37e6f68aac7be9cbbb4e90ca87a6e592c155eb35 /net
parent83803034f4233d810c4adc52008921da060c55d1 (diff)
[TCP]: Add TCP Westwood congestion control module.
This is the existing 2.6.12 Westwood code moved from tcp_input to the new congestion framework. A lot of the inline functions have been eliminated to try and make it clearer. Signed-off-by: Stephen Hemminger <shemminger@osdl.org> Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net')
-rw-r--r--net/ipv4/Kconfig15
-rw-r--r--net/ipv4/Makefile1
-rw-r--r--net/ipv4/tcp_westwood.c259
3 files changed, 275 insertions, 0 deletions
diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig
index 712ebacacb62..adbe855d931a 100644
--- a/net/ipv4/Kconfig
+++ b/net/ipv4/Kconfig
@@ -452,6 +452,21 @@ config TCP_CONG_BIC
452 increase provides TCP friendliness. 452 increase provides TCP friendliness.
453 See http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/ 453 See http://www.csc.ncsu.edu/faculty/rhee/export/bitcp/
454 454
455config TCP_CONG_WESTWOOD
456 tristate "TCP Westwood+"
457 depends on INET
458 default m
459 ---help---
460 TCP Westwood+ is a sender-side only modification of the TCP Reno
461 protocol stack that optimizes the performance of TCP congestion
462 control. It is based on end-to-end bandwidth estimation to set
463 congestion window and slow start threshold after a congestion
464 episode. Using this estimation, TCP Westwood+ adaptively sets a
465 slow start threshold and a congestion window which takes into
466 account the bandwidth used at the time congestion is experienced.
467 TCP Westwood+ significantly increases fairness wrt TCP Reno in
468 wired networks and throughput over wireless links.
469
455endmenu 470endmenu
456 471
457source "net/ipv4/ipvs/Kconfig" 472source "net/ipv4/ipvs/Kconfig"
diff --git a/net/ipv4/Makefile b/net/ipv4/Makefile
index 1d1cac5ac06a..dedfbe62a104 100644
--- a/net/ipv4/Makefile
+++ b/net/ipv4/Makefile
@@ -32,6 +32,7 @@ obj-$(CONFIG_IP_VS) += ipvs/
32obj-$(CONFIG_IP_TCPDIAG) += tcp_diag.o 32obj-$(CONFIG_IP_TCPDIAG) += tcp_diag.o
33obj-$(CONFIG_IP_ROUTE_MULTIPATH_CACHED) += multipath.o 33obj-$(CONFIG_IP_ROUTE_MULTIPATH_CACHED) += multipath.o
34obj-$(CONFIG_TCP_CONG_BIC) += tcp_bic.o 34obj-$(CONFIG_TCP_CONG_BIC) += tcp_bic.o
35obj-$(CONFIG_TCP_CONG_WESTWOOD) += tcp_westwood.o
35 36
36obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \ 37obj-$(CONFIG_XFRM) += xfrm4_policy.o xfrm4_state.o xfrm4_input.o \
37 xfrm4_output.o 38 xfrm4_output.o
diff --git a/net/ipv4/tcp_westwood.c b/net/ipv4/tcp_westwood.c
new file mode 100644
index 000000000000..ef827242c940
--- /dev/null
+++ b/net/ipv4/tcp_westwood.c
@@ -0,0 +1,259 @@
1/*
2 * TCP Westwood+
3 *
4 * Angelo Dell'Aera: TCP Westwood+ support
5 */
6
7#include <linux/config.h>
8#include <linux/mm.h>
9#include <linux/module.h>
10#include <linux/skbuff.h>
11#include <linux/tcp_diag.h>
12#include <net/tcp.h>
13
14/* TCP Westwood structure */
15struct westwood {
16 u32 bw_ns_est; /* first bandwidth estimation..not too smoothed 8) */
17 u32 bw_est; /* bandwidth estimate */
18 u32 rtt_win_sx; /* here starts a new evaluation... */
19 u32 bk;
20 u32 snd_una; /* used for evaluating the number of acked bytes */
21 u32 cumul_ack;
22 u32 accounted;
23 u32 rtt;
24 u32 rtt_min; /* minimum observed RTT */
25};
26
27
28/* TCP Westwood functions and constants */
29#define TCP_WESTWOOD_RTT_MIN (HZ/20) /* 50ms */
30#define TCP_WESTWOOD_INIT_RTT (20*HZ) /* maybe too conservative?! */
31
32/*
33 * @tcp_westwood_create
34 * This function initializes fields used in TCP Westwood+,
35 * it is called after the initial SYN, so the sequence numbers
36 * are correct but new passive connections we have no
37 * information about RTTmin at this time so we simply set it to
38 * TCP_WESTWOOD_INIT_RTT. This value was chosen to be too conservative
39 * since in this way we're sure it will be updated in a consistent
40 * way as soon as possible. It will reasonably happen within the first
41 * RTT period of the connection lifetime.
42 */
43static void tcp_westwood_init(struct tcp_sock *tp)
44{
45 struct westwood *w = tcp_ca(tp);
46
47 w->bk = 0;
48 w->bw_ns_est = 0;
49 w->bw_est = 0;
50 w->accounted = 0;
51 w->cumul_ack = 0;
52 w->rtt_min = w->rtt = TCP_WESTWOOD_INIT_RTT;
53 w->rtt_win_sx = tcp_time_stamp;
54 w->snd_una = tp->snd_una;
55}
56
57/*
58 * @westwood_do_filter
59 * Low-pass filter. Implemented using constant coefficients.
60 */
61static inline u32 westwood_do_filter(u32 a, u32 b)
62{
63 return (((7 * a) + b) >> 3);
64}
65
66static inline void westwood_filter(struct westwood *w, u32 delta)
67{
68 w->bw_ns_est = westwood_do_filter(w->bw_ns_est, w->bk / delta);
69 w->bw_est = westwood_do_filter(w->bw_est, w->bw_ns_est);
70}
71
72/*
73 * @westwood_pkts_acked
74 * Called after processing group of packets.
75 * but all westwood needs is the last sample of srtt.
76 */
77static void tcp_westwood_pkts_acked(struct tcp_sock *tp, u32 cnt)
78{
79 struct westwood *w = tcp_ca(tp);
80 if (cnt > 0)
81 w->rtt = tp->srtt >> 3;
82}
83
84/*
85 * @westwood_update_window
86 * It updates RTT evaluation window if it is the right moment to do
87 * it. If so it calls filter for evaluating bandwidth.
88 */
89static void westwood_update_window(struct tcp_sock *tp)
90{
91 struct westwood *w = tcp_ca(tp);
92 s32 delta = tcp_time_stamp - w->rtt_win_sx;
93
94 /*
95 * See if a RTT-window has passed.
96 * Be careful since if RTT is less than
97 * 50ms we don't filter but we continue 'building the sample'.
98 * This minimum limit was chosen since an estimation on small
99 * time intervals is better to avoid...
100 * Obviously on a LAN we reasonably will always have
101 * right_bound = left_bound + WESTWOOD_RTT_MIN
102 */
103 if (w->rtt && delta > max_t(u32, w->rtt, TCP_WESTWOOD_RTT_MIN)) {
104 westwood_filter(w, delta);
105
106 w->bk = 0;
107 w->rtt_win_sx = tcp_time_stamp;
108 }
109}
110
111/*
112 * @westwood_fast_bw
113 * It is called when we are in fast path. In particular it is called when
114 * header prediction is successful. In such case in fact update is
115 * straight forward and doesn't need any particular care.
116 */
117static inline void westwood_fast_bw(struct tcp_sock *tp)
118{
119 struct westwood *w = tcp_ca(tp);
120
121 westwood_update_window(tp);
122
123 w->bk += tp->snd_una - w->snd_una;
124 w->snd_una = tp->snd_una;
125 w->rtt_min = min(w->rtt, w->rtt_min);
126}
127
128/*
129 * @westwood_acked_count
130 * This function evaluates cumul_ack for evaluating bk in case of
131 * delayed or partial acks.
132 */
133static inline u32 westwood_acked_count(struct tcp_sock *tp)
134{
135 struct westwood *w = tcp_ca(tp);
136
137 w->cumul_ack = tp->snd_una - w->snd_una;
138
139 /* If cumul_ack is 0 this is a dupack since it's not moving
140 * tp->snd_una.
141 */
142 if (!w->cumul_ack) {
143 w->accounted += tp->mss_cache;
144 w->cumul_ack = tp->mss_cache;
145 }
146
147 if (w->cumul_ack > tp->mss_cache) {
148 /* Partial or delayed ack */
149 if (w->accounted >= w->cumul_ack) {
150 w->accounted -= w->cumul_ack;
151 w->cumul_ack = tp->mss_cache;
152 } else {
153 w->cumul_ack -= w->accounted;
154 w->accounted = 0;
155 }
156 }
157
158 w->snd_una = tp->snd_una;
159
160 return w->cumul_ack;
161}
162
163static inline u32 westwood_bw_rttmin(const struct tcp_sock *tp)
164{
165 struct westwood *w = tcp_ca(tp);
166 return max_t(u32, (w->bw_est * w->rtt_min) / tp->mss_cache, 2);
167}
168
169/*
170 * TCP Westwood
171 * Here limit is evaluated as Bw estimation*RTTmin (for obtaining it
172 * in packets we use mss_cache). Rttmin is guaranteed to be >= 2
173 * so avoids ever returning 0.
174 */
175static u32 tcp_westwood_cwnd_min(struct tcp_sock *tp)
176{
177 return westwood_bw_rttmin(tp);
178}
179
180static void tcp_westwood_event(struct tcp_sock *tp, enum tcp_ca_event event)
181{
182 struct westwood *w = tcp_ca(tp);
183
184 switch(event) {
185 case CA_EVENT_FAST_ACK:
186 westwood_fast_bw(tp);
187 break;
188
189 case CA_EVENT_COMPLETE_CWR:
190 tp->snd_cwnd = tp->snd_ssthresh = westwood_bw_rttmin(tp);
191 break;
192
193 case CA_EVENT_FRTO:
194 tp->snd_ssthresh = westwood_bw_rttmin(tp);
195 break;
196
197 case CA_EVENT_SLOW_ACK:
198 westwood_update_window(tp);
199 w->bk += westwood_acked_count(tp);
200 w->rtt_min = min(w->rtt, w->rtt_min);
201 break;
202
203 default:
204 /* don't care */
205 break;
206 }
207}
208
209
210/* Extract info for Tcp socket info provided via netlink. */
211static void tcp_westwood_info(struct tcp_sock *tp, u32 ext,
212 struct sk_buff *skb)
213{
214 const struct westwood *ca = tcp_ca(tp);
215 if (ext & (1<<(TCPDIAG_VEGASINFO-1))) {
216 struct rtattr *rta;
217 struct tcpvegas_info *info;
218
219 rta = __RTA_PUT(skb, TCPDIAG_VEGASINFO, sizeof(*info));
220 info = RTA_DATA(rta);
221 info->tcpv_enabled = 1;
222 info->tcpv_rttcnt = 0;
223 info->tcpv_rtt = jiffies_to_usecs(ca->rtt);
224 info->tcpv_minrtt = jiffies_to_usecs(ca->rtt_min);
225 rtattr_failure: ;
226 }
227}
228
229
230static struct tcp_congestion_ops tcp_westwood = {
231 .init = tcp_westwood_init,
232 .ssthresh = tcp_reno_ssthresh,
233 .cong_avoid = tcp_reno_cong_avoid,
234 .min_cwnd = tcp_westwood_cwnd_min,
235 .cwnd_event = tcp_westwood_event,
236 .get_info = tcp_westwood_info,
237 .pkts_acked = tcp_westwood_pkts_acked,
238
239 .owner = THIS_MODULE,
240 .name = "westwood"
241};
242
243static int __init tcp_westwood_register(void)
244{
245 BUG_ON(sizeof(struct westwood) > TCP_CA_PRIV_SIZE);
246 return tcp_register_congestion_control(&tcp_westwood);
247}
248
249static void __exit tcp_westwood_unregister(void)
250{
251 tcp_unregister_congestion_control(&tcp_westwood);
252}
253
254module_init(tcp_westwood_register);
255module_exit(tcp_westwood_unregister);
256
257MODULE_AUTHOR("Stephen Hemminger, Angelo Dell'Aera");
258MODULE_LICENSE("GPL");
259MODULE_DESCRIPTION("TCP Westwood+");