Loading...
1/*
2 * TCP Illinois congestion control.
3 * Home page:
4 * http://www.ews.uiuc.edu/~shaoliu/tcpillinois/index.html
5 *
6 * The algorithm is described in:
7 * "TCP-Illinois: A Loss and Delay-Based Congestion Control Algorithm
8 * for High-Speed Networks"
9 * http://www.ifp.illinois.edu/~srikant/Papers/liubassri06perf.pdf
10 *
11 * Implemented from description in paper and ns-2 simulation.
12 * Copyright (C) 2007 Stephen Hemminger <shemminger@linux-foundation.org>
13 */
14
15#include <linux/module.h>
16#include <linux/skbuff.h>
17#include <linux/inet_diag.h>
18#include <asm/div64.h>
19#include <net/tcp.h>
20
21#define ALPHA_SHIFT 7
22#define ALPHA_SCALE (1u<<ALPHA_SHIFT)
23#define ALPHA_MIN ((3*ALPHA_SCALE)/10) /* ~0.3 */
24#define ALPHA_MAX (10*ALPHA_SCALE) /* 10.0 */
25#define ALPHA_BASE ALPHA_SCALE /* 1.0 */
26#define U32_MAX ((u32)~0U)
27#define RTT_MAX (U32_MAX / ALPHA_MAX) /* 3.3 secs */
28
29#define BETA_SHIFT 6
30#define BETA_SCALE (1u<<BETA_SHIFT)
31#define BETA_MIN (BETA_SCALE/8) /* 0.125 */
32#define BETA_MAX (BETA_SCALE/2) /* 0.5 */
33#define BETA_BASE BETA_MAX
34
35static int win_thresh __read_mostly = 15;
36module_param(win_thresh, int, 0);
37MODULE_PARM_DESC(win_thresh, "Window threshold for starting adaptive sizing");
38
39static int theta __read_mostly = 5;
40module_param(theta, int, 0);
41MODULE_PARM_DESC(theta, "# of fast RTT's before full growth");
42
43/* TCP Illinois Parameters */
44struct illinois {
45 u64 sum_rtt; /* sum of rtt's measured within last rtt */
46 u16 cnt_rtt; /* # of rtts measured within last rtt */
47 u32 base_rtt; /* min of all rtt in usec */
48 u32 max_rtt; /* max of all rtt in usec */
49 u32 end_seq; /* right edge of current RTT */
50 u32 alpha; /* Additive increase */
51 u32 beta; /* Muliplicative decrease */
52 u16 acked; /* # packets acked by current ACK */
53 u8 rtt_above; /* average rtt has gone above threshold */
54 u8 rtt_low; /* # of rtts measurements below threshold */
55};
56
57static void rtt_reset(struct sock *sk)
58{
59 struct tcp_sock *tp = tcp_sk(sk);
60 struct illinois *ca = inet_csk_ca(sk);
61
62 ca->end_seq = tp->snd_nxt;
63 ca->cnt_rtt = 0;
64 ca->sum_rtt = 0;
65
66 /* TODO: age max_rtt? */
67}
68
69static void tcp_illinois_init(struct sock *sk)
70{
71 struct illinois *ca = inet_csk_ca(sk);
72
73 ca->alpha = ALPHA_MAX;
74 ca->beta = BETA_BASE;
75 ca->base_rtt = 0x7fffffff;
76 ca->max_rtt = 0;
77
78 ca->acked = 0;
79 ca->rtt_low = 0;
80 ca->rtt_above = 0;
81
82 rtt_reset(sk);
83}
84
85/* Measure RTT for each ack. */
86static void tcp_illinois_acked(struct sock *sk, u32 pkts_acked, s32 rtt)
87{
88 struct illinois *ca = inet_csk_ca(sk);
89
90 ca->acked = pkts_acked;
91
92 /* dup ack, no rtt sample */
93 if (rtt < 0)
94 return;
95
96 /* ignore bogus values, this prevents wraparound in alpha math */
97 if (rtt > RTT_MAX)
98 rtt = RTT_MAX;
99
100 /* keep track of minimum RTT seen so far */
101 if (ca->base_rtt > rtt)
102 ca->base_rtt = rtt;
103
104 /* and max */
105 if (ca->max_rtt < rtt)
106 ca->max_rtt = rtt;
107
108 ++ca->cnt_rtt;
109 ca->sum_rtt += rtt;
110}
111
112/* Maximum queuing delay */
113static inline u32 max_delay(const struct illinois *ca)
114{
115 return ca->max_rtt - ca->base_rtt;
116}
117
118/* Average queuing delay */
119static inline u32 avg_delay(const struct illinois *ca)
120{
121 u64 t = ca->sum_rtt;
122
123 do_div(t, ca->cnt_rtt);
124 return t - ca->base_rtt;
125}
126
127/*
128 * Compute value of alpha used for additive increase.
129 * If small window then use 1.0, equivalent to Reno.
130 *
131 * For larger windows, adjust based on average delay.
132 * A. If average delay is at minimum (we are uncongested),
133 * then use large alpha (10.0) to increase faster.
134 * B. If average delay is at maximum (getting congested)
135 * then use small alpha (0.3)
136 *
137 * The result is a convex window growth curve.
138 */
139static u32 alpha(struct illinois *ca, u32 da, u32 dm)
140{
141 u32 d1 = dm / 100; /* Low threshold */
142
143 if (da <= d1) {
144 /* If never got out of low delay zone, then use max */
145 if (!ca->rtt_above)
146 return ALPHA_MAX;
147
148 /* Wait for 5 good RTT's before allowing alpha to go alpha max.
149 * This prevents one good RTT from causing sudden window increase.
150 */
151 if (++ca->rtt_low < theta)
152 return ca->alpha;
153
154 ca->rtt_low = 0;
155 ca->rtt_above = 0;
156 return ALPHA_MAX;
157 }
158
159 ca->rtt_above = 1;
160
161 /*
162 * Based on:
163 *
164 * (dm - d1) amin amax
165 * k1 = -------------------
166 * amax - amin
167 *
168 * (dm - d1) amin
169 * k2 = ---------------- - d1
170 * amax - amin
171 *
172 * k1
173 * alpha = ----------
174 * k2 + da
175 */
176
177 dm -= d1;
178 da -= d1;
179 return (dm * ALPHA_MAX) /
180 (dm + (da * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN);
181}
182
183/*
184 * Beta used for multiplicative decrease.
185 * For small window sizes returns same value as Reno (0.5)
186 *
187 * If delay is small (10% of max) then beta = 1/8
188 * If delay is up to 80% of max then beta = 1/2
189 * In between is a linear function
190 */
191static u32 beta(u32 da, u32 dm)
192{
193 u32 d2, d3;
194
195 d2 = dm / 10;
196 if (da <= d2)
197 return BETA_MIN;
198
199 d3 = (8 * dm) / 10;
200 if (da >= d3 || d3 <= d2)
201 return BETA_MAX;
202
203 /*
204 * Based on:
205 *
206 * bmin d3 - bmax d2
207 * k3 = -------------------
208 * d3 - d2
209 *
210 * bmax - bmin
211 * k4 = -------------
212 * d3 - d2
213 *
214 * b = k3 + k4 da
215 */
216 return (BETA_MIN * d3 - BETA_MAX * d2 + (BETA_MAX - BETA_MIN) * da)
217 / (d3 - d2);
218}
219
220/* Update alpha and beta values once per RTT */
221static void update_params(struct sock *sk)
222{
223 struct tcp_sock *tp = tcp_sk(sk);
224 struct illinois *ca = inet_csk_ca(sk);
225
226 if (tp->snd_cwnd < win_thresh) {
227 ca->alpha = ALPHA_BASE;
228 ca->beta = BETA_BASE;
229 } else if (ca->cnt_rtt > 0) {
230 u32 dm = max_delay(ca);
231 u32 da = avg_delay(ca);
232
233 ca->alpha = alpha(ca, da, dm);
234 ca->beta = beta(da, dm);
235 }
236
237 rtt_reset(sk);
238}
239
240/*
241 * In case of loss, reset to default values
242 */
243static void tcp_illinois_state(struct sock *sk, u8 new_state)
244{
245 struct illinois *ca = inet_csk_ca(sk);
246
247 if (new_state == TCP_CA_Loss) {
248 ca->alpha = ALPHA_BASE;
249 ca->beta = BETA_BASE;
250 ca->rtt_low = 0;
251 ca->rtt_above = 0;
252 rtt_reset(sk);
253 }
254}
255
256/*
257 * Increase window in response to successful acknowledgment.
258 */
259static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 in_flight)
260{
261 struct tcp_sock *tp = tcp_sk(sk);
262 struct illinois *ca = inet_csk_ca(sk);
263
264 if (after(ack, ca->end_seq))
265 update_params(sk);
266
267 /* RFC2861 only increase cwnd if fully utilized */
268 if (!tcp_is_cwnd_limited(sk, in_flight))
269 return;
270
271 /* In slow start */
272 if (tp->snd_cwnd <= tp->snd_ssthresh)
273 tcp_slow_start(tp);
274
275 else {
276 u32 delta;
277
278 /* snd_cwnd_cnt is # of packets since last cwnd increment */
279 tp->snd_cwnd_cnt += ca->acked;
280 ca->acked = 1;
281
282 /* This is close approximation of:
283 * tp->snd_cwnd += alpha/tp->snd_cwnd
284 */
285 delta = (tp->snd_cwnd_cnt * ca->alpha) >> ALPHA_SHIFT;
286 if (delta >= tp->snd_cwnd) {
287 tp->snd_cwnd = min(tp->snd_cwnd + delta / tp->snd_cwnd,
288 (u32) tp->snd_cwnd_clamp);
289 tp->snd_cwnd_cnt = 0;
290 }
291 }
292}
293
294static u32 tcp_illinois_ssthresh(struct sock *sk)
295{
296 struct tcp_sock *tp = tcp_sk(sk);
297 struct illinois *ca = inet_csk_ca(sk);
298
299 /* Multiplicative decrease */
300 return max(tp->snd_cwnd - ((tp->snd_cwnd * ca->beta) >> BETA_SHIFT), 2U);
301}
302
303
304/* Extract info for Tcp socket info provided via netlink. */
305static void tcp_illinois_info(struct sock *sk, u32 ext,
306 struct sk_buff *skb)
307{
308 const struct illinois *ca = inet_csk_ca(sk);
309
310 if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
311 struct tcpvegas_info info = {
312 .tcpv_enabled = 1,
313 .tcpv_rttcnt = ca->cnt_rtt,
314 .tcpv_minrtt = ca->base_rtt,
315 };
316 u64 t = ca->sum_rtt;
317
318 do_div(t, ca->cnt_rtt);
319 info.tcpv_rtt = t;
320
321 nla_put(skb, INET_DIAG_VEGASINFO, sizeof(info), &info);
322 }
323}
324
325static struct tcp_congestion_ops tcp_illinois __read_mostly = {
326 .flags = TCP_CONG_RTT_STAMP,
327 .init = tcp_illinois_init,
328 .ssthresh = tcp_illinois_ssthresh,
329 .min_cwnd = tcp_reno_min_cwnd,
330 .cong_avoid = tcp_illinois_cong_avoid,
331 .set_state = tcp_illinois_state,
332 .get_info = tcp_illinois_info,
333 .pkts_acked = tcp_illinois_acked,
334
335 .owner = THIS_MODULE,
336 .name = "illinois",
337};
338
339static int __init tcp_illinois_register(void)
340{
341 BUILD_BUG_ON(sizeof(struct illinois) > ICSK_CA_PRIV_SIZE);
342 return tcp_register_congestion_control(&tcp_illinois);
343}
344
345static void __exit tcp_illinois_unregister(void)
346{
347 tcp_unregister_congestion_control(&tcp_illinois);
348}
349
350module_init(tcp_illinois_register);
351module_exit(tcp_illinois_unregister);
352
353MODULE_AUTHOR("Stephen Hemminger, Shao Liu");
354MODULE_LICENSE("GPL");
355MODULE_DESCRIPTION("TCP Illinois");
356MODULE_VERSION("1.0");
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * TCP Illinois congestion control.
4 * Home page:
5 * http://www.ews.uiuc.edu/~shaoliu/tcpillinois/index.html
6 *
7 * The algorithm is described in:
8 * "TCP-Illinois: A Loss and Delay-Based Congestion Control Algorithm
9 * for High-Speed Networks"
10 * http://tamerbasar.csl.illinois.edu/LiuBasarSrikantPerfEvalArtJun2008.pdf
11 *
12 * Implemented from description in paper and ns-2 simulation.
13 * Copyright (C) 2007 Stephen Hemminger <shemminger@linux-foundation.org>
14 */
15
16#include <linux/module.h>
17#include <linux/skbuff.h>
18#include <linux/inet_diag.h>
19#include <asm/div64.h>
20#include <net/tcp.h>
21
22#define ALPHA_SHIFT 7
23#define ALPHA_SCALE (1u<<ALPHA_SHIFT)
24#define ALPHA_MIN ((3*ALPHA_SCALE)/10) /* ~0.3 */
25#define ALPHA_MAX (10*ALPHA_SCALE) /* 10.0 */
26#define ALPHA_BASE ALPHA_SCALE /* 1.0 */
27#define RTT_MAX (U32_MAX / ALPHA_MAX) /* 3.3 secs */
28
29#define BETA_SHIFT 6
30#define BETA_SCALE (1u<<BETA_SHIFT)
31#define BETA_MIN (BETA_SCALE/8) /* 0.125 */
32#define BETA_MAX (BETA_SCALE/2) /* 0.5 */
33#define BETA_BASE BETA_MAX
34
35static int win_thresh __read_mostly = 15;
36module_param(win_thresh, int, 0);
37MODULE_PARM_DESC(win_thresh, "Window threshold for starting adaptive sizing");
38
39static int theta __read_mostly = 5;
40module_param(theta, int, 0);
41MODULE_PARM_DESC(theta, "# of fast RTT's before full growth");
42
43/* TCP Illinois Parameters */
44struct illinois {
45 u64 sum_rtt; /* sum of rtt's measured within last rtt */
46 u16 cnt_rtt; /* # of rtts measured within last rtt */
47 u32 base_rtt; /* min of all rtt in usec */
48 u32 max_rtt; /* max of all rtt in usec */
49 u32 end_seq; /* right edge of current RTT */
50 u32 alpha; /* Additive increase */
51 u32 beta; /* Muliplicative decrease */
52 u16 acked; /* # packets acked by current ACK */
53 u8 rtt_above; /* average rtt has gone above threshold */
54 u8 rtt_low; /* # of rtts measurements below threshold */
55};
56
57static void rtt_reset(struct sock *sk)
58{
59 struct tcp_sock *tp = tcp_sk(sk);
60 struct illinois *ca = inet_csk_ca(sk);
61
62 ca->end_seq = tp->snd_nxt;
63 ca->cnt_rtt = 0;
64 ca->sum_rtt = 0;
65
66 /* TODO: age max_rtt? */
67}
68
69static void tcp_illinois_init(struct sock *sk)
70{
71 struct illinois *ca = inet_csk_ca(sk);
72
73 ca->alpha = ALPHA_MAX;
74 ca->beta = BETA_BASE;
75 ca->base_rtt = 0x7fffffff;
76 ca->max_rtt = 0;
77
78 ca->acked = 0;
79 ca->rtt_low = 0;
80 ca->rtt_above = 0;
81
82 rtt_reset(sk);
83}
84
85/* Measure RTT for each ack. */
86static void tcp_illinois_acked(struct sock *sk, const struct ack_sample *sample)
87{
88 struct illinois *ca = inet_csk_ca(sk);
89 s32 rtt_us = sample->rtt_us;
90
91 ca->acked = sample->pkts_acked;
92
93 /* dup ack, no rtt sample */
94 if (rtt_us < 0)
95 return;
96
97 /* ignore bogus values, this prevents wraparound in alpha math */
98 if (rtt_us > RTT_MAX)
99 rtt_us = RTT_MAX;
100
101 /* keep track of minimum RTT seen so far */
102 if (ca->base_rtt > rtt_us)
103 ca->base_rtt = rtt_us;
104
105 /* and max */
106 if (ca->max_rtt < rtt_us)
107 ca->max_rtt = rtt_us;
108
109 ++ca->cnt_rtt;
110 ca->sum_rtt += rtt_us;
111}
112
113/* Maximum queuing delay */
114static inline u32 max_delay(const struct illinois *ca)
115{
116 return ca->max_rtt - ca->base_rtt;
117}
118
119/* Average queuing delay */
120static inline u32 avg_delay(const struct illinois *ca)
121{
122 u64 t = ca->sum_rtt;
123
124 do_div(t, ca->cnt_rtt);
125 return t - ca->base_rtt;
126}
127
128/*
129 * Compute value of alpha used for additive increase.
130 * If small window then use 1.0, equivalent to Reno.
131 *
132 * For larger windows, adjust based on average delay.
133 * A. If average delay is at minimum (we are uncongested),
134 * then use large alpha (10.0) to increase faster.
135 * B. If average delay is at maximum (getting congested)
136 * then use small alpha (0.3)
137 *
138 * The result is a convex window growth curve.
139 */
140static u32 alpha(struct illinois *ca, u32 da, u32 dm)
141{
142 u32 d1 = dm / 100; /* Low threshold */
143
144 if (da <= d1) {
145 /* If never got out of low delay zone, then use max */
146 if (!ca->rtt_above)
147 return ALPHA_MAX;
148
149 /* Wait for 5 good RTT's before allowing alpha to go alpha max.
150 * This prevents one good RTT from causing sudden window increase.
151 */
152 if (++ca->rtt_low < theta)
153 return ca->alpha;
154
155 ca->rtt_low = 0;
156 ca->rtt_above = 0;
157 return ALPHA_MAX;
158 }
159
160 ca->rtt_above = 1;
161
162 /*
163 * Based on:
164 *
165 * (dm - d1) amin amax
166 * k1 = -------------------
167 * amax - amin
168 *
169 * (dm - d1) amin
170 * k2 = ---------------- - d1
171 * amax - amin
172 *
173 * k1
174 * alpha = ----------
175 * k2 + da
176 */
177
178 dm -= d1;
179 da -= d1;
180 return (dm * ALPHA_MAX) /
181 (dm + (da * (ALPHA_MAX - ALPHA_MIN)) / ALPHA_MIN);
182}
183
184/*
185 * Beta used for multiplicative decrease.
186 * For small window sizes returns same value as Reno (0.5)
187 *
188 * If delay is small (10% of max) then beta = 1/8
189 * If delay is up to 80% of max then beta = 1/2
190 * In between is a linear function
191 */
192static u32 beta(u32 da, u32 dm)
193{
194 u32 d2, d3;
195
196 d2 = dm / 10;
197 if (da <= d2)
198 return BETA_MIN;
199
200 d3 = (8 * dm) / 10;
201 if (da >= d3 || d3 <= d2)
202 return BETA_MAX;
203
204 /*
205 * Based on:
206 *
207 * bmin d3 - bmax d2
208 * k3 = -------------------
209 * d3 - d2
210 *
211 * bmax - bmin
212 * k4 = -------------
213 * d3 - d2
214 *
215 * b = k3 + k4 da
216 */
217 return (BETA_MIN * d3 - BETA_MAX * d2 + (BETA_MAX - BETA_MIN) * da)
218 / (d3 - d2);
219}
220
221/* Update alpha and beta values once per RTT */
222static void update_params(struct sock *sk)
223{
224 struct tcp_sock *tp = tcp_sk(sk);
225 struct illinois *ca = inet_csk_ca(sk);
226
227 if (tcp_snd_cwnd(tp) < win_thresh) {
228 ca->alpha = ALPHA_BASE;
229 ca->beta = BETA_BASE;
230 } else if (ca->cnt_rtt > 0) {
231 u32 dm = max_delay(ca);
232 u32 da = avg_delay(ca);
233
234 ca->alpha = alpha(ca, da, dm);
235 ca->beta = beta(da, dm);
236 }
237
238 rtt_reset(sk);
239}
240
241/*
242 * In case of loss, reset to default values
243 */
244static void tcp_illinois_state(struct sock *sk, u8 new_state)
245{
246 struct illinois *ca = inet_csk_ca(sk);
247
248 if (new_state == TCP_CA_Loss) {
249 ca->alpha = ALPHA_BASE;
250 ca->beta = BETA_BASE;
251 ca->rtt_low = 0;
252 ca->rtt_above = 0;
253 rtt_reset(sk);
254 }
255}
256
257/*
258 * Increase window in response to successful acknowledgment.
259 */
260static void tcp_illinois_cong_avoid(struct sock *sk, u32 ack, u32 acked)
261{
262 struct tcp_sock *tp = tcp_sk(sk);
263 struct illinois *ca = inet_csk_ca(sk);
264
265 if (after(ack, ca->end_seq))
266 update_params(sk);
267
268 /* RFC2861 only increase cwnd if fully utilized */
269 if (!tcp_is_cwnd_limited(sk))
270 return;
271
272 /* In slow start */
273 if (tcp_in_slow_start(tp))
274 tcp_slow_start(tp, acked);
275
276 else {
277 u32 delta;
278
279 /* snd_cwnd_cnt is # of packets since last cwnd increment */
280 tp->snd_cwnd_cnt += ca->acked;
281 ca->acked = 1;
282
283 /* This is close approximation of:
284 * tp->snd_cwnd += alpha/tp->snd_cwnd
285 */
286 delta = (tp->snd_cwnd_cnt * ca->alpha) >> ALPHA_SHIFT;
287 if (delta >= tcp_snd_cwnd(tp)) {
288 tcp_snd_cwnd_set(tp, min(tcp_snd_cwnd(tp) + delta / tcp_snd_cwnd(tp),
289 (u32)tp->snd_cwnd_clamp));
290 tp->snd_cwnd_cnt = 0;
291 }
292 }
293}
294
295static u32 tcp_illinois_ssthresh(struct sock *sk)
296{
297 struct tcp_sock *tp = tcp_sk(sk);
298 struct illinois *ca = inet_csk_ca(sk);
299 u32 decr;
300
301 /* Multiplicative decrease */
302 decr = (tcp_snd_cwnd(tp) * ca->beta) >> BETA_SHIFT;
303 return max(tcp_snd_cwnd(tp) - decr, 2U);
304}
305
306/* Extract info for Tcp socket info provided via netlink. */
307static size_t tcp_illinois_info(struct sock *sk, u32 ext, int *attr,
308 union tcp_cc_info *info)
309{
310 const struct illinois *ca = inet_csk_ca(sk);
311
312 if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
313 info->vegas.tcpv_enabled = 1;
314 info->vegas.tcpv_rttcnt = ca->cnt_rtt;
315 info->vegas.tcpv_minrtt = ca->base_rtt;
316 info->vegas.tcpv_rtt = 0;
317
318 if (info->vegas.tcpv_rttcnt > 0) {
319 u64 t = ca->sum_rtt;
320
321 do_div(t, info->vegas.tcpv_rttcnt);
322 info->vegas.tcpv_rtt = t;
323 }
324 *attr = INET_DIAG_VEGASINFO;
325 return sizeof(struct tcpvegas_info);
326 }
327 return 0;
328}
329
330static struct tcp_congestion_ops tcp_illinois __read_mostly = {
331 .init = tcp_illinois_init,
332 .ssthresh = tcp_illinois_ssthresh,
333 .undo_cwnd = tcp_reno_undo_cwnd,
334 .cong_avoid = tcp_illinois_cong_avoid,
335 .set_state = tcp_illinois_state,
336 .get_info = tcp_illinois_info,
337 .pkts_acked = tcp_illinois_acked,
338
339 .owner = THIS_MODULE,
340 .name = "illinois",
341};
342
343static int __init tcp_illinois_register(void)
344{
345 BUILD_BUG_ON(sizeof(struct illinois) > ICSK_CA_PRIV_SIZE);
346 return tcp_register_congestion_control(&tcp_illinois);
347}
348
349static void __exit tcp_illinois_unregister(void)
350{
351 tcp_unregister_congestion_control(&tcp_illinois);
352}
353
354module_init(tcp_illinois_register);
355module_exit(tcp_illinois_unregister);
356
357MODULE_AUTHOR("Stephen Hemminger, Shao Liu");
358MODULE_LICENSE("GPL");
359MODULE_DESCRIPTION("TCP Illinois");
360MODULE_VERSION("1.0");