xref: /openbmc/linux/net/ipv4/tcp_cubic.c (revision 8b235f2f)
1 /*
2  * TCP CUBIC: Binary Increase Congestion control for TCP v2.3
3  * Home page:
4  *      http://netsrv.csc.ncsu.edu/twiki/bin/view/Main/BIC
5  * This is from the implementation of CUBIC TCP in
6  * Sangtae Ha, Injong Rhee and Lisong Xu,
7  *  "CUBIC: A New TCP-Friendly High-Speed TCP Variant"
8  *  in ACM SIGOPS Operating System Review, July 2008.
9  * Available from:
10  *  http://netsrv.csc.ncsu.edu/export/cubic_a_new_tcp_2008.pdf
11  *
12  * CUBIC integrates a new slow start algorithm, called HyStart.
13  * The details of HyStart are presented in
14  *  Sangtae Ha and Injong Rhee,
15  *  "Taming the Elephants: New TCP Slow Start", NCSU TechReport 2008.
16  * Available from:
17  *  http://netsrv.csc.ncsu.edu/export/hystart_techreport_2008.pdf
18  *
19  * All testing results are available from:
20  * http://netsrv.csc.ncsu.edu/wiki/index.php/TCP_Testing
21  *
22  * Unless CUBIC is enabled and congestion window is large
23  * this behaves the same as the original Reno.
24  */
25 
26 #include <linux/mm.h>
27 #include <linux/module.h>
28 #include <linux/math64.h>
29 #include <net/tcp.h>
30 
31 #define BICTCP_BETA_SCALE    1024	/* Scale factor beta calculation
32 					 * max_cwnd = snd_cwnd * beta
33 					 */
34 #define	BICTCP_HZ		10	/* BIC HZ 2^10 = 1024 */
35 
36 /* Two methods of hybrid slow start */
37 #define HYSTART_ACK_TRAIN	0x1
38 #define HYSTART_DELAY		0x2
39 
40 /* Number of delay samples for detecting the increase of delay */
41 #define HYSTART_MIN_SAMPLES	8
42 #define HYSTART_DELAY_MIN	(4U<<3)
43 #define HYSTART_DELAY_MAX	(16U<<3)
44 #define HYSTART_DELAY_THRESH(x)	clamp(x, HYSTART_DELAY_MIN, HYSTART_DELAY_MAX)
45 
46 static int fast_convergence __read_mostly = 1;
47 static int beta __read_mostly = 717;	/* = 717/1024 (BICTCP_BETA_SCALE) */
48 static int initial_ssthresh __read_mostly;
49 static int bic_scale __read_mostly = 41;
50 static int tcp_friendliness __read_mostly = 1;
51 
52 static int hystart __read_mostly = 1;
53 static int hystart_detect __read_mostly = HYSTART_ACK_TRAIN | HYSTART_DELAY;
54 static int hystart_low_window __read_mostly = 16;
55 static int hystart_ack_delta __read_mostly = 2;
56 
57 static u32 cube_rtt_scale __read_mostly;
58 static u32 beta_scale __read_mostly;
59 static u64 cube_factor __read_mostly;
60 
61 /* Note parameters that are used for precomputing scale factors are read-only */
62 module_param(fast_convergence, int, 0644);
63 MODULE_PARM_DESC(fast_convergence, "turn on/off fast convergence");
64 module_param(beta, int, 0644);
65 MODULE_PARM_DESC(beta, "beta for multiplicative increase");
66 module_param(initial_ssthresh, int, 0644);
67 MODULE_PARM_DESC(initial_ssthresh, "initial value of slow start threshold");
68 module_param(bic_scale, int, 0444);
69 MODULE_PARM_DESC(bic_scale, "scale (scaled by 1024) value for bic function (bic_scale/1024)");
70 module_param(tcp_friendliness, int, 0644);
71 MODULE_PARM_DESC(tcp_friendliness, "turn on/off tcp friendliness");
72 module_param(hystart, int, 0644);
73 MODULE_PARM_DESC(hystart, "turn on/off hybrid slow start algorithm");
74 module_param(hystart_detect, int, 0644);
75 MODULE_PARM_DESC(hystart_detect, "hyrbrid slow start detection mechanisms"
76 		 " 1: packet-train 2: delay 3: both packet-train and delay");
77 module_param(hystart_low_window, int, 0644);
78 MODULE_PARM_DESC(hystart_low_window, "lower bound cwnd for hybrid slow start");
79 module_param(hystart_ack_delta, int, 0644);
80 MODULE_PARM_DESC(hystart_ack_delta, "spacing between ack's indicating train (msecs)");
81 
82 /* BIC TCP Parameters */
83 struct bictcp {
84 	u32	cnt;		/* increase cwnd by 1 after ACKs */
85 	u32	last_max_cwnd;	/* last maximum snd_cwnd */
86 	u32	loss_cwnd;	/* congestion window at last loss */
87 	u32	last_cwnd;	/* the last snd_cwnd */
88 	u32	last_time;	/* time when updated last_cwnd */
89 	u32	bic_origin_point;/* origin point of bic function */
90 	u32	bic_K;		/* time to origin point
91 				   from the beginning of the current epoch */
92 	u32	delay_min;	/* min delay (msec << 3) */
93 	u32	epoch_start;	/* beginning of an epoch */
94 	u32	ack_cnt;	/* number of acks */
95 	u32	tcp_cwnd;	/* estimated tcp cwnd */
96 	u16	unused;
97 	u8	sample_cnt;	/* number of samples to decide curr_rtt */
98 	u8	found;		/* the exit point is found? */
99 	u32	round_start;	/* beginning of each round */
100 	u32	end_seq;	/* end_seq of the round */
101 	u32	last_ack;	/* last time when the ACK spacing is close */
102 	u32	curr_rtt;	/* the minimum rtt of current round */
103 };
104 
105 static inline void bictcp_reset(struct bictcp *ca)
106 {
107 	ca->cnt = 0;
108 	ca->last_max_cwnd = 0;
109 	ca->last_cwnd = 0;
110 	ca->last_time = 0;
111 	ca->bic_origin_point = 0;
112 	ca->bic_K = 0;
113 	ca->delay_min = 0;
114 	ca->epoch_start = 0;
115 	ca->ack_cnt = 0;
116 	ca->tcp_cwnd = 0;
117 	ca->found = 0;
118 }
119 
120 static inline u32 bictcp_clock(void)
121 {
122 #if HZ < 1000
123 	return ktime_to_ms(ktime_get_real());
124 #else
125 	return jiffies_to_msecs(jiffies);
126 #endif
127 }
128 
129 static inline void bictcp_hystart_reset(struct sock *sk)
130 {
131 	struct tcp_sock *tp = tcp_sk(sk);
132 	struct bictcp *ca = inet_csk_ca(sk);
133 
134 	ca->round_start = ca->last_ack = bictcp_clock();
135 	ca->end_seq = tp->snd_nxt;
136 	ca->curr_rtt = 0;
137 	ca->sample_cnt = 0;
138 }
139 
140 static void bictcp_init(struct sock *sk)
141 {
142 	struct bictcp *ca = inet_csk_ca(sk);
143 
144 	bictcp_reset(ca);
145 	ca->loss_cwnd = 0;
146 
147 	if (hystart)
148 		bictcp_hystart_reset(sk);
149 
150 	if (!hystart && initial_ssthresh)
151 		tcp_sk(sk)->snd_ssthresh = initial_ssthresh;
152 }
153 
154 /* calculate the cubic root of x using a table lookup followed by one
155  * Newton-Raphson iteration.
156  * Avg err ~= 0.195%
157  */
158 static u32 cubic_root(u64 a)
159 {
160 	u32 x, b, shift;
161 	/*
162 	 * cbrt(x) MSB values for x MSB values in [0..63].
163 	 * Precomputed then refined by hand - Willy Tarreau
164 	 *
165 	 * For x in [0..63],
166 	 *   v = cbrt(x << 18) - 1
167 	 *   cbrt(x) = (v[x] + 10) >> 6
168 	 */
169 	static const u8 v[] = {
170 		/* 0x00 */    0,   54,   54,   54,  118,  118,  118,  118,
171 		/* 0x08 */  123,  129,  134,  138,  143,  147,  151,  156,
172 		/* 0x10 */  157,  161,  164,  168,  170,  173,  176,  179,
173 		/* 0x18 */  181,  185,  187,  190,  192,  194,  197,  199,
174 		/* 0x20 */  200,  202,  204,  206,  209,  211,  213,  215,
175 		/* 0x28 */  217,  219,  221,  222,  224,  225,  227,  229,
176 		/* 0x30 */  231,  232,  234,  236,  237,  239,  240,  242,
177 		/* 0x38 */  244,  245,  246,  248,  250,  251,  252,  254,
178 	};
179 
180 	b = fls64(a);
181 	if (b < 7) {
182 		/* a in [0..63] */
183 		return ((u32)v[(u32)a] + 35) >> 6;
184 	}
185 
186 	b = ((b * 84) >> 8) - 1;
187 	shift = (a >> (b * 3));
188 
189 	x = ((u32)(((u32)v[shift] + 10) << b)) >> 6;
190 
191 	/*
192 	 * Newton-Raphson iteration
193 	 *                         2
194 	 * x    = ( 2 * x  +  a / x  ) / 3
195 	 *  k+1          k         k
196 	 */
197 	x = (2 * x + (u32)div64_u64(a, (u64)x * (u64)(x - 1)));
198 	x = ((x * 341) >> 10);
199 	return x;
200 }
201 
202 /*
203  * Compute congestion window to use.
204  */
205 static inline void bictcp_update(struct bictcp *ca, u32 cwnd, u32 acked)
206 {
207 	u32 delta, bic_target, max_cnt;
208 	u64 offs, t;
209 
210 	ca->ack_cnt += acked;	/* count the number of ACKed packets */
211 
212 	if (ca->last_cwnd == cwnd &&
213 	    (s32)(tcp_time_stamp - ca->last_time) <= HZ / 32)
214 		return;
215 
216 	/* The CUBIC function can update ca->cnt at most once per jiffy.
217 	 * On all cwnd reduction events, ca->epoch_start is set to 0,
218 	 * which will force a recalculation of ca->cnt.
219 	 */
220 	if (ca->epoch_start && tcp_time_stamp == ca->last_time)
221 		goto tcp_friendliness;
222 
223 	ca->last_cwnd = cwnd;
224 	ca->last_time = tcp_time_stamp;
225 
226 	if (ca->epoch_start == 0) {
227 		ca->epoch_start = tcp_time_stamp;	/* record beginning */
228 		ca->ack_cnt = acked;			/* start counting */
229 		ca->tcp_cwnd = cwnd;			/* syn with cubic */
230 
231 		if (ca->last_max_cwnd <= cwnd) {
232 			ca->bic_K = 0;
233 			ca->bic_origin_point = cwnd;
234 		} else {
235 			/* Compute new K based on
236 			 * (wmax-cwnd) * (srtt>>3 / HZ) / c * 2^(3*bictcp_HZ)
237 			 */
238 			ca->bic_K = cubic_root(cube_factor
239 					       * (ca->last_max_cwnd - cwnd));
240 			ca->bic_origin_point = ca->last_max_cwnd;
241 		}
242 	}
243 
244 	/* cubic function - calc*/
245 	/* calculate c * time^3 / rtt,
246 	 *  while considering overflow in calculation of time^3
247 	 * (so time^3 is done by using 64 bit)
248 	 * and without the support of division of 64bit numbers
249 	 * (so all divisions are done by using 32 bit)
250 	 *  also NOTE the unit of those veriables
251 	 *	  time  = (t - K) / 2^bictcp_HZ
252 	 *	  c = bic_scale >> 10
253 	 * rtt  = (srtt >> 3) / HZ
254 	 * !!! The following code does not have overflow problems,
255 	 * if the cwnd < 1 million packets !!!
256 	 */
257 
258 	t = (s32)(tcp_time_stamp - ca->epoch_start);
259 	t += msecs_to_jiffies(ca->delay_min >> 3);
260 	/* change the unit from HZ to bictcp_HZ */
261 	t <<= BICTCP_HZ;
262 	do_div(t, HZ);
263 
264 	if (t < ca->bic_K)		/* t - K */
265 		offs = ca->bic_K - t;
266 	else
267 		offs = t - ca->bic_K;
268 
269 	/* c/rtt * (t-K)^3 */
270 	delta = (cube_rtt_scale * offs * offs * offs) >> (10+3*BICTCP_HZ);
271 	if (t < ca->bic_K)                            /* below origin*/
272 		bic_target = ca->bic_origin_point - delta;
273 	else                                          /* above origin*/
274 		bic_target = ca->bic_origin_point + delta;
275 
276 	/* cubic function - calc bictcp_cnt*/
277 	if (bic_target > cwnd) {
278 		ca->cnt = cwnd / (bic_target - cwnd);
279 	} else {
280 		ca->cnt = 100 * cwnd;              /* very small increment*/
281 	}
282 
283 	/*
284 	 * The initial growth of cubic function may be too conservative
285 	 * when the available bandwidth is still unknown.
286 	 */
287 	if (ca->last_max_cwnd == 0 && ca->cnt > 20)
288 		ca->cnt = 20;	/* increase cwnd 5% per RTT */
289 
290 tcp_friendliness:
291 	/* TCP Friendly */
292 	if (tcp_friendliness) {
293 		u32 scale = beta_scale;
294 
295 		delta = (cwnd * scale) >> 3;
296 		while (ca->ack_cnt > delta) {		/* update tcp cwnd */
297 			ca->ack_cnt -= delta;
298 			ca->tcp_cwnd++;
299 		}
300 
301 		if (ca->tcp_cwnd > cwnd) {	/* if bic is slower than tcp */
302 			delta = ca->tcp_cwnd - cwnd;
303 			max_cnt = cwnd / delta;
304 			if (ca->cnt > max_cnt)
305 				ca->cnt = max_cnt;
306 		}
307 	}
308 
309 	/* The maximum rate of cwnd increase CUBIC allows is 1 packet per
310 	 * 2 packets ACKed, meaning cwnd grows at 1.5x per RTT.
311 	 */
312 	ca->cnt = max(ca->cnt, 2U);
313 }
314 
315 static void bictcp_cong_avoid(struct sock *sk, u32 ack, u32 acked)
316 {
317 	struct tcp_sock *tp = tcp_sk(sk);
318 	struct bictcp *ca = inet_csk_ca(sk);
319 
320 	if (!tcp_is_cwnd_limited(sk))
321 		return;
322 
323 	if (tcp_in_slow_start(tp)) {
324 		if (hystart && after(ack, ca->end_seq))
325 			bictcp_hystart_reset(sk);
326 		acked = tcp_slow_start(tp, acked);
327 		if (!acked)
328 			return;
329 	}
330 	bictcp_update(ca, tp->snd_cwnd, acked);
331 	tcp_cong_avoid_ai(tp, ca->cnt, acked);
332 }
333 
334 static u32 bictcp_recalc_ssthresh(struct sock *sk)
335 {
336 	const struct tcp_sock *tp = tcp_sk(sk);
337 	struct bictcp *ca = inet_csk_ca(sk);
338 
339 	ca->epoch_start = 0;	/* end of epoch */
340 
341 	/* Wmax and fast convergence */
342 	if (tp->snd_cwnd < ca->last_max_cwnd && fast_convergence)
343 		ca->last_max_cwnd = (tp->snd_cwnd * (BICTCP_BETA_SCALE + beta))
344 			/ (2 * BICTCP_BETA_SCALE);
345 	else
346 		ca->last_max_cwnd = tp->snd_cwnd;
347 
348 	ca->loss_cwnd = tp->snd_cwnd;
349 
350 	return max((tp->snd_cwnd * beta) / BICTCP_BETA_SCALE, 2U);
351 }
352 
353 static u32 bictcp_undo_cwnd(struct sock *sk)
354 {
355 	struct bictcp *ca = inet_csk_ca(sk);
356 
357 	return max(tcp_sk(sk)->snd_cwnd, ca->loss_cwnd);
358 }
359 
360 static void bictcp_state(struct sock *sk, u8 new_state)
361 {
362 	if (new_state == TCP_CA_Loss) {
363 		bictcp_reset(inet_csk_ca(sk));
364 		bictcp_hystart_reset(sk);
365 	}
366 }
367 
368 static void hystart_update(struct sock *sk, u32 delay)
369 {
370 	struct tcp_sock *tp = tcp_sk(sk);
371 	struct bictcp *ca = inet_csk_ca(sk);
372 
373 	if (ca->found & hystart_detect)
374 		return;
375 
376 	if (hystart_detect & HYSTART_ACK_TRAIN) {
377 		u32 now = bictcp_clock();
378 
379 		/* first detection parameter - ack-train detection */
380 		if ((s32)(now - ca->last_ack) <= hystart_ack_delta) {
381 			ca->last_ack = now;
382 			if ((s32)(now - ca->round_start) > ca->delay_min >> 4) {
383 				ca->found |= HYSTART_ACK_TRAIN;
384 				NET_INC_STATS_BH(sock_net(sk),
385 						 LINUX_MIB_TCPHYSTARTTRAINDETECT);
386 				NET_ADD_STATS_BH(sock_net(sk),
387 						 LINUX_MIB_TCPHYSTARTTRAINCWND,
388 						 tp->snd_cwnd);
389 				tp->snd_ssthresh = tp->snd_cwnd;
390 			}
391 		}
392 	}
393 
394 	if (hystart_detect & HYSTART_DELAY) {
395 		/* obtain the minimum delay of more than sampling packets */
396 		if (ca->sample_cnt < HYSTART_MIN_SAMPLES) {
397 			if (ca->curr_rtt == 0 || ca->curr_rtt > delay)
398 				ca->curr_rtt = delay;
399 
400 			ca->sample_cnt++;
401 		} else {
402 			if (ca->curr_rtt > ca->delay_min +
403 			    HYSTART_DELAY_THRESH(ca->delay_min >> 3)) {
404 				ca->found |= HYSTART_DELAY;
405 				NET_INC_STATS_BH(sock_net(sk),
406 						 LINUX_MIB_TCPHYSTARTDELAYDETECT);
407 				NET_ADD_STATS_BH(sock_net(sk),
408 						 LINUX_MIB_TCPHYSTARTDELAYCWND,
409 						 tp->snd_cwnd);
410 				tp->snd_ssthresh = tp->snd_cwnd;
411 			}
412 		}
413 	}
414 }
415 
416 /* Track delayed acknowledgment ratio using sliding window
417  * ratio = (15*ratio + sample) / 16
418  */
419 static void bictcp_acked(struct sock *sk, u32 cnt, s32 rtt_us)
420 {
421 	const struct tcp_sock *tp = tcp_sk(sk);
422 	struct bictcp *ca = inet_csk_ca(sk);
423 	u32 delay;
424 
425 	/* Some calls are for duplicates without timetamps */
426 	if (rtt_us < 0)
427 		return;
428 
429 	/* Discard delay samples right after fast recovery */
430 	if (ca->epoch_start && (s32)(tcp_time_stamp - ca->epoch_start) < HZ)
431 		return;
432 
433 	delay = (rtt_us << 3) / USEC_PER_MSEC;
434 	if (delay == 0)
435 		delay = 1;
436 
437 	/* first time call or link delay decreases */
438 	if (ca->delay_min == 0 || ca->delay_min > delay)
439 		ca->delay_min = delay;
440 
441 	/* hystart triggers when cwnd is larger than some threshold */
442 	if (hystart && tcp_in_slow_start(tp) &&
443 	    tp->snd_cwnd >= hystart_low_window)
444 		hystart_update(sk, delay);
445 }
446 
447 static struct tcp_congestion_ops cubictcp __read_mostly = {
448 	.init		= bictcp_init,
449 	.ssthresh	= bictcp_recalc_ssthresh,
450 	.cong_avoid	= bictcp_cong_avoid,
451 	.set_state	= bictcp_state,
452 	.undo_cwnd	= bictcp_undo_cwnd,
453 	.pkts_acked     = bictcp_acked,
454 	.owner		= THIS_MODULE,
455 	.name		= "cubic",
456 };
457 
458 static int __init cubictcp_register(void)
459 {
460 	BUILD_BUG_ON(sizeof(struct bictcp) > ICSK_CA_PRIV_SIZE);
461 
462 	/* Precompute a bunch of the scaling factors that are used per-packet
463 	 * based on SRTT of 100ms
464 	 */
465 
466 	beta_scale = 8*(BICTCP_BETA_SCALE+beta) / 3
467 		/ (BICTCP_BETA_SCALE - beta);
468 
469 	cube_rtt_scale = (bic_scale * 10);	/* 1024*c/rtt */
470 
471 	/* calculate the "K" for (wmax-cwnd) = c/rtt * K^3
472 	 *  so K = cubic_root( (wmax-cwnd)*rtt/c )
473 	 * the unit of K is bictcp_HZ=2^10, not HZ
474 	 *
475 	 *  c = bic_scale >> 10
476 	 *  rtt = 100ms
477 	 *
478 	 * the following code has been designed and tested for
479 	 * cwnd < 1 million packets
480 	 * RTT < 100 seconds
481 	 * HZ < 1,000,00  (corresponding to 10 nano-second)
482 	 */
483 
484 	/* 1/c * 2^2*bictcp_HZ * srtt */
485 	cube_factor = 1ull << (10+3*BICTCP_HZ); /* 2^40 */
486 
487 	/* divide by bic_scale and by constant Srtt (100ms) */
488 	do_div(cube_factor, bic_scale * 10);
489 
490 	return tcp_register_congestion_control(&cubictcp);
491 }
492 
493 static void __exit cubictcp_unregister(void)
494 {
495 	tcp_unregister_congestion_control(&cubictcp);
496 }
497 
498 module_init(cubictcp_register);
499 module_exit(cubictcp_unregister);
500 
501 MODULE_AUTHOR("Sangtae Ha, Stephen Hemminger");
502 MODULE_LICENSE("GPL");
503 MODULE_DESCRIPTION("CUBIC TCP");
504 MODULE_VERSION("2.3");
505