xref: /openbmc/linux/net/ipv4/tcp_nv.c (revision ca481398)
1 /*
2  * TCP NV: TCP with Congestion Avoidance
3  *
4  * TCP-NV is a successor of TCP-Vegas that has been developed to
5  * deal with the issues that occur in modern networks.
6  * Like TCP-Vegas, TCP-NV supports true congestion avoidance,
7  * the ability to detect congestion before packet losses occur.
8  * When congestion (queue buildup) starts to occur, TCP-NV
9  * predicts what the cwnd size should be for the current
10  * throughput and it reduces the cwnd proportionally to
11  * the difference between the current cwnd and the predicted cwnd.
12  *
13  * NV is only recommeneded for traffic within a data center, and when
14  * all the flows are NV (at least those within the data center). This
15  * is due to the inherent unfairness between flows using losses to
16  * detect congestion (congestion control) and those that use queue
17  * buildup to detect congestion (congestion avoidance).
18  *
19  * Note: High NIC coalescence values may lower the performance of NV
20  * due to the increased noise in RTT values. In particular, we have
21  * seen issues with rx-frames values greater than 8.
22  *
23  * TODO:
24  * 1) Add mechanism to deal with reverse congestion.
25  */
26 
27 #include <linux/mm.h>
28 #include <linux/module.h>
29 #include <linux/math64.h>
30 #include <net/tcp.h>
31 #include <linux/inet_diag.h>
32 
33 /* TCP NV parameters
34  *
35  * nv_pad		Max number of queued packets allowed in network
36  * nv_pad_buffer	Do not grow cwnd if this closed to nv_pad
37  * nv_reset_period	How often (in) seconds)to reset min_rtt
38  * nv_min_cwnd		Don't decrease cwnd below this if there are no losses
39  * nv_cong_dec_mult	Decrease cwnd by X% (30%) of congestion when detected
40  * nv_ssthresh_factor	On congestion set ssthresh to this * <desired cwnd> / 8
41  * nv_rtt_factor	RTT averaging factor
42  * nv_loss_dec_factor	Decrease cwnd by this (50%) when losses occur
43  * nv_dec_eval_min_calls	Wait this many RTT measurements before dec cwnd
44  * nv_inc_eval_min_calls	Wait this many RTT measurements before inc cwnd
45  * nv_ssthresh_eval_min_calls	Wait this many RTT measurements before stopping
46  *				slow-start due to congestion
47  * nv_stop_rtt_cnt	Only grow cwnd for this many RTTs after non-congestion
48  * nv_rtt_min_cnt	Wait these many RTTs before making congesion decision
49  * nv_cwnd_growth_rate_neg
50  * nv_cwnd_growth_rate_pos
51  *	How quickly to double growth rate (not rate) of cwnd when not
52  *	congested. One value (nv_cwnd_growth_rate_neg) for when
53  *	rate < 1 pkt/RTT (after losses). The other (nv_cwnd_growth_rate_pos)
54  *	otherwise.
55  */
56 
57 static int nv_pad __read_mostly = 10;
58 static int nv_pad_buffer __read_mostly = 2;
59 static int nv_reset_period __read_mostly = 5; /* in seconds */
60 static int nv_min_cwnd __read_mostly = 2;
61 static int nv_cong_dec_mult __read_mostly = 30 * 128 / 100; /* = 30% */
62 static int nv_ssthresh_factor __read_mostly = 8; /* = 1 */
63 static int nv_rtt_factor __read_mostly = 128; /* = 1/2*old + 1/2*new */
64 static int nv_loss_dec_factor __read_mostly = 512; /* => 50% */
65 static int nv_cwnd_growth_rate_neg __read_mostly = 8;
66 static int nv_cwnd_growth_rate_pos __read_mostly; /* 0 => fixed like Reno */
67 static int nv_dec_eval_min_calls __read_mostly = 60;
68 static int nv_inc_eval_min_calls __read_mostly = 20;
69 static int nv_ssthresh_eval_min_calls __read_mostly = 30;
70 static int nv_stop_rtt_cnt __read_mostly = 10;
71 static int nv_rtt_min_cnt __read_mostly = 2;
72 
73 module_param(nv_pad, int, 0644);
74 MODULE_PARM_DESC(nv_pad, "max queued packets allowed in network");
75 module_param(nv_reset_period, int, 0644);
76 MODULE_PARM_DESC(nv_reset_period, "nv_min_rtt reset period (secs)");
77 module_param(nv_min_cwnd, int, 0644);
78 MODULE_PARM_DESC(nv_min_cwnd, "NV will not decrease cwnd below this value"
79 		 " without losses");
80 
81 /* TCP NV Parameters */
82 struct tcpnv {
83 	unsigned long nv_min_rtt_reset_jiffies;  /* when to switch to
84 						  * nv_min_rtt_new */
85 	s8  cwnd_growth_factor;	/* Current cwnd growth factor,
86 				 * < 0 => less than 1 packet/RTT */
87 	u8  available8;
88 	u16 available16;
89 	u8  nv_allow_cwnd_growth:1, /* whether cwnd can grow */
90 		nv_reset:1,	    /* whether to reset values */
91 		nv_catchup:1;	    /* whether we are growing because
92 				     * of temporary cwnd decrease */
93 	u8  nv_eval_call_cnt;	/* call count since last eval */
94 	u8  nv_min_cwnd;	/* nv won't make a ca decision if cwnd is
95 				 * smaller than this. It may grow to handle
96 				 * TSO, LRO and interrupt coalescence because
97 				 * with these a small cwnd cannot saturate
98 				 * the link. Note that this is different from
99 				 * the file local nv_min_cwnd */
100 	u8  nv_rtt_cnt;		/* RTTs without making ca decision */;
101 	u32 nv_last_rtt;	/* last rtt */
102 	u32 nv_min_rtt;		/* active min rtt. Used to determine slope */
103 	u32 nv_min_rtt_new;	/* min rtt for future use */
104 	u32 nv_rtt_max_rate;	/* max rate seen during current RTT */
105 	u32 nv_rtt_start_seq;	/* current RTT ends when packet arrives
106 				 * acking beyond nv_rtt_start_seq */
107 	u32 nv_last_snd_una;	/* Previous value of tp->snd_una. It is
108 				 * used to determine bytes acked since last
109 				 * call to bictcp_acked */
110 	u32 nv_no_cong_cnt;	/* Consecutive no congestion decisions */
111 };
112 
113 #define NV_INIT_RTT	  U32_MAX
114 #define NV_MIN_CWND	  4
115 #define NV_MIN_CWND_GROW  2
116 #define NV_TSO_CWND_BOUND 80
117 
118 static inline void tcpnv_reset(struct tcpnv *ca, struct sock *sk)
119 {
120 	struct tcp_sock *tp = tcp_sk(sk);
121 
122 	ca->nv_reset = 0;
123 	ca->nv_no_cong_cnt = 0;
124 	ca->nv_rtt_cnt = 0;
125 	ca->nv_last_rtt = 0;
126 	ca->nv_rtt_max_rate = 0;
127 	ca->nv_rtt_start_seq = tp->snd_una;
128 	ca->nv_eval_call_cnt = 0;
129 	ca->nv_last_snd_una = tp->snd_una;
130 }
131 
132 static void tcpnv_init(struct sock *sk)
133 {
134 	struct tcpnv *ca = inet_csk_ca(sk);
135 
136 	tcpnv_reset(ca, sk);
137 
138 	ca->nv_allow_cwnd_growth = 1;
139 	ca->nv_min_rtt_reset_jiffies = jiffies + 2 * HZ;
140 	ca->nv_min_rtt = NV_INIT_RTT;
141 	ca->nv_min_rtt_new = NV_INIT_RTT;
142 	ca->nv_min_cwnd = NV_MIN_CWND;
143 	ca->nv_catchup = 0;
144 	ca->cwnd_growth_factor = 0;
145 }
146 
147 static void tcpnv_cong_avoid(struct sock *sk, u32 ack, u32 acked)
148 {
149 	struct tcp_sock *tp = tcp_sk(sk);
150 	struct tcpnv *ca = inet_csk_ca(sk);
151 	u32 cnt;
152 
153 	if (!tcp_is_cwnd_limited(sk))
154 		return;
155 
156 	/* Only grow cwnd if NV has not detected congestion */
157 	if (!ca->nv_allow_cwnd_growth)
158 		return;
159 
160 	if (tcp_in_slow_start(tp)) {
161 		acked = tcp_slow_start(tp, acked);
162 		if (!acked)
163 			return;
164 	}
165 
166 	if (ca->cwnd_growth_factor < 0) {
167 		cnt = tp->snd_cwnd << -ca->cwnd_growth_factor;
168 		tcp_cong_avoid_ai(tp, cnt, acked);
169 	} else {
170 		cnt = max(4U, tp->snd_cwnd >> ca->cwnd_growth_factor);
171 		tcp_cong_avoid_ai(tp, cnt, acked);
172 	}
173 }
174 
175 static u32 tcpnv_recalc_ssthresh(struct sock *sk)
176 {
177 	const struct tcp_sock *tp = tcp_sk(sk);
178 
179 	return max((tp->snd_cwnd * nv_loss_dec_factor) >> 10, 2U);
180 }
181 
182 static void tcpnv_state(struct sock *sk, u8 new_state)
183 {
184 	struct tcpnv *ca = inet_csk_ca(sk);
185 
186 	if (new_state == TCP_CA_Open && ca->nv_reset) {
187 		tcpnv_reset(ca, sk);
188 	} else if (new_state == TCP_CA_Loss || new_state == TCP_CA_CWR ||
189 		new_state == TCP_CA_Recovery) {
190 		ca->nv_reset = 1;
191 		ca->nv_allow_cwnd_growth = 0;
192 		if (new_state == TCP_CA_Loss) {
193 			/* Reset cwnd growth factor to Reno value */
194 			if (ca->cwnd_growth_factor > 0)
195 				ca->cwnd_growth_factor = 0;
196 			/* Decrease growth rate if allowed */
197 			if (nv_cwnd_growth_rate_neg > 0 &&
198 			    ca->cwnd_growth_factor > -8)
199 				ca->cwnd_growth_factor--;
200 		}
201 	}
202 }
203 
204 /* Do congestion avoidance calculations for TCP-NV
205  */
206 static void tcpnv_acked(struct sock *sk, const struct ack_sample *sample)
207 {
208 	const struct inet_connection_sock *icsk = inet_csk(sk);
209 	struct tcp_sock *tp = tcp_sk(sk);
210 	struct tcpnv *ca = inet_csk_ca(sk);
211 	unsigned long now = jiffies;
212 	s64 rate64 = 0;
213 	u32 rate, max_win, cwnd_by_slope;
214 	u32 avg_rtt;
215 	u32 bytes_acked = 0;
216 
217 	/* Some calls are for duplicates without timetamps */
218 	if (sample->rtt_us < 0)
219 		return;
220 
221 	/* If not in TCP_CA_Open or TCP_CA_Disorder states, skip. */
222 	if (icsk->icsk_ca_state != TCP_CA_Open &&
223 	    icsk->icsk_ca_state != TCP_CA_Disorder)
224 		return;
225 
226 	/* Stop cwnd growth if we were in catch up mode */
227 	if (ca->nv_catchup && tp->snd_cwnd >= nv_min_cwnd) {
228 		ca->nv_catchup = 0;
229 		ca->nv_allow_cwnd_growth = 0;
230 	}
231 
232 	bytes_acked = tp->snd_una - ca->nv_last_snd_una;
233 	ca->nv_last_snd_una = tp->snd_una;
234 
235 	if (sample->in_flight == 0)
236 		return;
237 
238 	/* Calculate moving average of RTT */
239 	if (nv_rtt_factor > 0) {
240 		if (ca->nv_last_rtt > 0) {
241 			avg_rtt = (((u64)sample->rtt_us) * nv_rtt_factor +
242 				   ((u64)ca->nv_last_rtt)
243 				   * (256 - nv_rtt_factor)) >> 8;
244 		} else {
245 			avg_rtt = sample->rtt_us;
246 			ca->nv_min_rtt = avg_rtt << 1;
247 		}
248 		ca->nv_last_rtt = avg_rtt;
249 	} else {
250 		avg_rtt = sample->rtt_us;
251 	}
252 
253 	/* rate in 100's bits per second */
254 	rate64 = ((u64)sample->in_flight) * 8000000;
255 	rate = (u32)div64_u64(rate64, (u64)(avg_rtt * 100));
256 
257 	/* Remember the maximum rate seen during this RTT
258 	 * Note: It may be more than one RTT. This function should be
259 	 *       called at least nv_dec_eval_min_calls times.
260 	 */
261 	if (ca->nv_rtt_max_rate < rate)
262 		ca->nv_rtt_max_rate = rate;
263 
264 	/* We have valid information, increment counter */
265 	if (ca->nv_eval_call_cnt < 255)
266 		ca->nv_eval_call_cnt++;
267 
268 	/* update min rtt if necessary */
269 	if (avg_rtt < ca->nv_min_rtt)
270 		ca->nv_min_rtt = avg_rtt;
271 
272 	/* update future min_rtt if necessary */
273 	if (avg_rtt < ca->nv_min_rtt_new)
274 		ca->nv_min_rtt_new = avg_rtt;
275 
276 	/* nv_min_rtt is updated with the minimum (possibley averaged) rtt
277 	 * seen in the last sysctl_tcp_nv_reset_period seconds (i.e. a
278 	 * warm reset). This new nv_min_rtt will be continued to be updated
279 	 * and be used for another sysctl_tcp_nv_reset_period seconds,
280 	 * when it will be updated again.
281 	 * In practice we introduce some randomness, so the actual period used
282 	 * is chosen randomly from the range:
283 	 *   [sysctl_tcp_nv_reset_period*3/4, sysctl_tcp_nv_reset_period*5/4)
284 	 */
285 	if (time_after_eq(now, ca->nv_min_rtt_reset_jiffies)) {
286 		unsigned char rand;
287 
288 		ca->nv_min_rtt = ca->nv_min_rtt_new;
289 		ca->nv_min_rtt_new = NV_INIT_RTT;
290 		get_random_bytes(&rand, 1);
291 		ca->nv_min_rtt_reset_jiffies =
292 			now + ((nv_reset_period * (384 + rand) * HZ) >> 9);
293 		/* Every so often we decrease ca->nv_min_cwnd in case previous
294 		 *  value is no longer accurate.
295 		 */
296 		ca->nv_min_cwnd = max(ca->nv_min_cwnd / 2, NV_MIN_CWND);
297 	}
298 
299 	/* Once per RTT check if we need to do congestion avoidance */
300 	if (before(ca->nv_rtt_start_seq, tp->snd_una)) {
301 		ca->nv_rtt_start_seq = tp->snd_nxt;
302 		if (ca->nv_rtt_cnt < 0xff)
303 			/* Increase counter for RTTs without CA decision */
304 			ca->nv_rtt_cnt++;
305 
306 		/* If this function is only called once within an RTT
307 		 * the cwnd is probably too small (in some cases due to
308 		 * tso, lro or interrupt coalescence), so we increase
309 		 * ca->nv_min_cwnd.
310 		 */
311 		if (ca->nv_eval_call_cnt == 1 &&
312 		    bytes_acked >= (ca->nv_min_cwnd - 1) * tp->mss_cache &&
313 		    ca->nv_min_cwnd < (NV_TSO_CWND_BOUND + 1)) {
314 			ca->nv_min_cwnd = min(ca->nv_min_cwnd
315 					      + NV_MIN_CWND_GROW,
316 					      NV_TSO_CWND_BOUND + 1);
317 			ca->nv_rtt_start_seq = tp->snd_nxt +
318 				ca->nv_min_cwnd * tp->mss_cache;
319 			ca->nv_eval_call_cnt = 0;
320 			ca->nv_allow_cwnd_growth = 1;
321 			return;
322 		}
323 
324 		/* Find the ideal cwnd for current rate from slope
325 		 * slope = 80000.0 * mss / nv_min_rtt
326 		 * cwnd_by_slope = nv_rtt_max_rate / slope
327 		 */
328 		cwnd_by_slope = (u32)
329 			div64_u64(((u64)ca->nv_rtt_max_rate) * ca->nv_min_rtt,
330 				  (u64)(80000 * tp->mss_cache));
331 		max_win = cwnd_by_slope + nv_pad;
332 
333 		/* If cwnd > max_win, decrease cwnd
334 		 * if cwnd < max_win, grow cwnd
335 		 * else leave the same
336 		 */
337 		if (tp->snd_cwnd > max_win) {
338 			/* there is congestion, check that it is ok
339 			 * to make a CA decision
340 			 * 1. We should have at least nv_dec_eval_min_calls
341 			 *    data points before making a CA  decision
342 			 * 2. We only make a congesion decision after
343 			 *    nv_rtt_min_cnt RTTs
344 			 */
345 			if (ca->nv_rtt_cnt < nv_rtt_min_cnt) {
346 				return;
347 			} else if (tp->snd_ssthresh == TCP_INFINITE_SSTHRESH) {
348 				if (ca->nv_eval_call_cnt <
349 				    nv_ssthresh_eval_min_calls)
350 					return;
351 				/* otherwise we will decrease cwnd */
352 			} else if (ca->nv_eval_call_cnt <
353 				   nv_dec_eval_min_calls) {
354 				if (ca->nv_allow_cwnd_growth &&
355 				    ca->nv_rtt_cnt > nv_stop_rtt_cnt)
356 					ca->nv_allow_cwnd_growth = 0;
357 				return;
358 			}
359 
360 			/* We have enough data to determine we are congested */
361 			ca->nv_allow_cwnd_growth = 0;
362 			tp->snd_ssthresh =
363 				(nv_ssthresh_factor * max_win) >> 3;
364 			if (tp->snd_cwnd - max_win > 2) {
365 				/* gap > 2, we do exponential cwnd decrease */
366 				int dec;
367 
368 				dec = max(2U, ((tp->snd_cwnd - max_win) *
369 					       nv_cong_dec_mult) >> 7);
370 				tp->snd_cwnd -= dec;
371 			} else if (nv_cong_dec_mult > 0) {
372 				tp->snd_cwnd = max_win;
373 			}
374 			if (ca->cwnd_growth_factor > 0)
375 				ca->cwnd_growth_factor = 0;
376 			ca->nv_no_cong_cnt = 0;
377 		} else if (tp->snd_cwnd <= max_win - nv_pad_buffer) {
378 			/* There is no congestion, grow cwnd if allowed*/
379 			if (ca->nv_eval_call_cnt < nv_inc_eval_min_calls)
380 				return;
381 
382 			ca->nv_allow_cwnd_growth = 1;
383 			ca->nv_no_cong_cnt++;
384 			if (ca->cwnd_growth_factor < 0 &&
385 			    nv_cwnd_growth_rate_neg > 0 &&
386 			    ca->nv_no_cong_cnt > nv_cwnd_growth_rate_neg) {
387 				ca->cwnd_growth_factor++;
388 				ca->nv_no_cong_cnt = 0;
389 			} else if (ca->cwnd_growth_factor >= 0 &&
390 				   nv_cwnd_growth_rate_pos > 0 &&
391 				   ca->nv_no_cong_cnt >
392 				   nv_cwnd_growth_rate_pos) {
393 				ca->cwnd_growth_factor++;
394 				ca->nv_no_cong_cnt = 0;
395 			}
396 		} else {
397 			/* cwnd is in-between, so do nothing */
398 			return;
399 		}
400 
401 		/* update state */
402 		ca->nv_eval_call_cnt = 0;
403 		ca->nv_rtt_cnt = 0;
404 		ca->nv_rtt_max_rate = 0;
405 
406 		/* Don't want to make cwnd < nv_min_cwnd
407 		 * (it wasn't before, if it is now is because nv
408 		 *  decreased it).
409 		 */
410 		if (tp->snd_cwnd < nv_min_cwnd)
411 			tp->snd_cwnd = nv_min_cwnd;
412 	}
413 }
414 
415 /* Extract info for Tcp socket info provided via netlink */
416 static size_t tcpnv_get_info(struct sock *sk, u32 ext, int *attr,
417 			     union tcp_cc_info *info)
418 {
419 	const struct tcpnv *ca = inet_csk_ca(sk);
420 
421 	if (ext & (1 << (INET_DIAG_VEGASINFO - 1))) {
422 		info->vegas.tcpv_enabled = 1;
423 		info->vegas.tcpv_rttcnt = ca->nv_rtt_cnt;
424 		info->vegas.tcpv_rtt = ca->nv_last_rtt;
425 		info->vegas.tcpv_minrtt = ca->nv_min_rtt;
426 
427 		*attr = INET_DIAG_VEGASINFO;
428 		return sizeof(struct tcpvegas_info);
429 	}
430 	return 0;
431 }
432 
433 static struct tcp_congestion_ops tcpnv __read_mostly = {
434 	.init		= tcpnv_init,
435 	.ssthresh	= tcpnv_recalc_ssthresh,
436 	.cong_avoid	= tcpnv_cong_avoid,
437 	.set_state	= tcpnv_state,
438 	.undo_cwnd	= tcp_reno_undo_cwnd,
439 	.pkts_acked     = tcpnv_acked,
440 	.get_info	= tcpnv_get_info,
441 
442 	.owner		= THIS_MODULE,
443 	.name		= "nv",
444 };
445 
446 static int __init tcpnv_register(void)
447 {
448 	BUILD_BUG_ON(sizeof(struct tcpnv) > ICSK_CA_PRIV_SIZE);
449 
450 	return tcp_register_congestion_control(&tcpnv);
451 }
452 
453 static void __exit tcpnv_unregister(void)
454 {
455 	tcp_unregister_congestion_control(&tcpnv);
456 }
457 
458 module_init(tcpnv_register);
459 module_exit(tcpnv_unregister);
460 
461 MODULE_AUTHOR("Lawrence Brakmo");
462 MODULE_LICENSE("GPL");
463 MODULE_DESCRIPTION("TCP NV");
464 MODULE_VERSION("1.0");
465