1 /*- 2 * Copyright (c) 2009-2013 3 * Swinburne University of Technology, Melbourne, Australia 4 * All rights reserved. 5 * 6 * This software was developed at the Centre for Advanced Internet 7 * Architectures, Swinburne University of Technology, by David Hayes, made 8 * possible in part by a gift from The Cisco University Research Program Fund, 9 * a corporate advised fund of Silicon Valley Community Foundation. Development 10 * and testing were further assisted by a grant from the FreeBSD Foundation. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * CAIA Delay-Gradient (CDG) congestion control algorithm 36 * 37 * An implemention of the delay-gradient congestion control algorithm proposed 38 * in the following paper: 39 * 40 * D. A. Hayes and G. Armitage, "Revisiting TCP Congestion Control using Delay 41 * Gradients", in IFIP Networking, Valencia, Spain, 9-13 May 2011. 42 * 43 * Developed as part of the NewTCP research project at Swinburne University of 44 * Technology's Centre for Advanced Internet Architectures, Melbourne, 45 * Australia. More details are available at: 46 * http://caia.swin.edu.au/urp/newtcp/ 47 */ 48 49 #include <sys/cdefs.h> 50 __FBSDID("$FreeBSD$"); 51 52 #include <sys/param.h> 53 #include <sys/hhook.h> 54 #include <sys/kernel.h> 55 #include <sys/khelp.h> 56 #include <sys/limits.h> 57 #include <sys/lock.h> 58 #include <sys/malloc.h> 59 #include <sys/module.h> 60 #include <sys/queue.h> 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/sysctl.h> 64 #include <sys/systm.h> 65 66 #include <net/vnet.h> 67 68 #include <netinet/tcp.h> 69 #include <netinet/tcp_seq.h> 70 #include <netinet/tcp_timer.h> 71 #include <netinet/tcp_var.h> 72 #include <netinet/cc/cc.h> 73 #include <netinet/cc/cc_module.h> 74 75 #include <netinet/khelp/h_ertt.h> 76 77 #include <vm/uma.h> 78 79 #define CDG_VERSION "0.1" 80 81 #define CAST_PTR_INT(X) (*((int*)(X))) 82 83 /* Private delay-gradient induced congestion control signal. */ 84 #define CC_CDG_DELAY 0x01000000 85 86 /* NewReno window deflation factor on loss (as a percentage). */ 87 #define RENO_BETA 50 88 89 /* Queue states. */ 90 #define CDG_Q_EMPTY 1 91 #define CDG_Q_RISING 2 92 #define CDG_Q_FALLING 3 93 #define CDG_Q_FULL 4 94 #define CDG_Q_UNKNOWN 9999 95 96 /* Number of bit shifts used in probexp lookup table. */ 97 #define EXP_PREC 15 98 99 /* Largest gradient represented in probexp lookup table. */ 100 #define MAXGRAD 5 101 102 /* 103 * Delay Precision Enhance - number of bit shifts used for qtrend related 104 * integer arithmetic precision. 105 */ 106 #define D_P_E 7 107 108 struct qdiff_sample { 109 long qdiff; 110 STAILQ_ENTRY(qdiff_sample) qdiff_lnk; 111 }; 112 113 struct cdg { 114 long max_qtrend; 115 long min_qtrend; 116 STAILQ_HEAD(minrtts_head, qdiff_sample) qdiffmin_q; 117 STAILQ_HEAD(maxrtts_head, qdiff_sample) qdiffmax_q; 118 long window_incr; 119 /* rttcount for window increase when in congestion avoidance */ 120 long rtt_count; 121 /* maximum measured rtt within an rtt period */ 122 int maxrtt_in_rtt; 123 /* maximum measured rtt within prev rtt period */ 124 int maxrtt_in_prevrtt; 125 /* minimum measured rtt within an rtt period */ 126 int minrtt_in_rtt; 127 /* minimum measured rtt within prev rtt period */ 128 int minrtt_in_prevrtt; 129 /* consecutive congestion episode counter */ 130 uint32_t consec_cong_cnt; 131 /* when tracking a new reno type loss window */ 132 uint32_t shadow_w; 133 /* maximum number of samples in the moving average queue */ 134 int sample_q_size; 135 /* number of samples in the moving average queue */ 136 int num_samples; 137 /* estimate of the queue state of the path */ 138 int queue_state; 139 }; 140 141 /* 142 * Lookup table for: 143 * (1 - exp(-x)) << EXP_PREC, where x = [0,MAXGRAD] in 2^-7 increments 144 * 145 * Note: probexp[0] is set to 10 (not 0) as a safety for very low increase 146 * gradients. 147 */ 148 static const int probexp[641] = { 149 10,255,508,759,1008,1255,1501,1744,1985,2225,2463,2698,2932,3165,3395,3624, 150 3850,4075,4299,4520,4740,4958,5175,5389,5602,5814,6024,6232,6438,6643,6846, 151 7048,7248,7447,7644,7839,8033,8226,8417,8606,8794,8981,9166,9350,9532,9713, 152 9892,10070,10247,10422,10596,10769,10940,11110,11278,11445,11611,11776,11939, 153 12101,12262,12422,12580,12737,12893,13048,13201,13354,13505,13655,13803,13951, 154 14097,14243,14387,14530,14672,14813,14952,15091,15229,15365,15500,15635,15768, 155 15900,16032,16162,16291,16419,16547,16673,16798,16922,17046,17168,17289,17410, 156 17529,17648,17766,17882,17998,18113,18227,18340,18453,18564,18675,18784,18893, 157 19001,19108,19215,19320,19425,19529,19632,19734,19835,19936,20036,20135,20233, 158 20331,20427,20523,20619,20713,20807,20900,20993,21084,21175,21265,21355,21444, 159 21532,21619,21706,21792,21878,21962,22046,22130,22213,22295,22376,22457,22537, 160 22617,22696,22774,22852,22929,23006,23082,23157,23232,23306,23380,23453,23525, 161 23597,23669,23739,23810,23879,23949,24017,24085,24153,24220,24286,24352,24418, 162 24483,24547,24611,24675,24738,24800,24862,24924,24985,25045,25106,25165,25224, 163 25283,25341,25399,25456,25513,25570,25626,25681,25737,25791,25846,25899,25953, 164 26006,26059,26111,26163,26214,26265,26316,26366,26416,26465,26514,26563,26611, 165 26659,26707,26754,26801,26847,26893,26939,26984,27029,27074,27118,27162,27206, 166 27249,27292,27335,27377,27419,27460,27502,27543,27583,27624,27664,27703,27743, 167 27782,27821,27859,27897,27935,27973,28010,28047,28084,28121,28157,28193,28228, 168 28263,28299,28333,28368,28402,28436,28470,28503,28536,28569,28602,28634,28667, 169 28699,28730,28762,28793,28824,28854,28885,28915,28945,28975,29004,29034,29063, 170 29092,29120,29149,29177,29205,29232,29260,29287,29314,29341,29368,29394,29421, 171 29447,29472,29498,29524,29549,29574,29599,29623,29648,29672,29696,29720,29744, 172 29767,29791,29814,29837,29860,29882,29905,29927,29949,29971,29993,30014,30036, 173 30057,30078,30099,30120,30141,30161,30181,30201,30221,30241,30261,30280,30300, 174 30319,30338,30357,30376,30394,30413,30431,30449,30467,30485,30503,30521,30538, 175 30555,30573,30590,30607,30624,30640,30657,30673,30690,30706,30722,30738,30753, 176 30769,30785,30800,30815,30831,30846,30861,30876,30890,30905,30919,30934,30948, 177 30962,30976,30990,31004,31018,31031,31045,31058,31072,31085,31098,31111,31124, 178 31137,31149,31162,31174,31187,31199,31211,31223,31235,31247,31259,31271,31283, 179 31294,31306,31317,31328,31339,31351,31362,31373,31383,31394,31405,31416,31426, 180 31436,31447,31457,31467,31477,31487,31497,31507,31517,31527,31537,31546,31556, 181 31565,31574,31584,31593,31602,31611,31620,31629,31638,31647,31655,31664,31673, 182 31681,31690,31698,31706,31715,31723,31731,31739,31747,31755,31763,31771,31778, 183 31786,31794,31801,31809,31816,31824,31831,31838,31846,31853,31860,31867,31874, 184 31881,31888,31895,31902,31908,31915,31922,31928,31935,31941,31948,31954,31960, 185 31967,31973,31979,31985,31991,31997,32003,32009,32015,32021,32027,32033,32038, 186 32044,32050,32055,32061,32066,32072,32077,32083,32088,32093,32098,32104,32109, 187 32114,32119,32124,32129,32134,32139,32144,32149,32154,32158,32163,32168,32173, 188 32177,32182,32186,32191,32195,32200,32204,32209,32213,32217,32222,32226,32230, 189 32234,32238,32242,32247,32251,32255,32259,32263,32267,32270,32274,32278,32282, 190 32286,32290,32293,32297,32301,32304,32308,32311,32315,32318,32322,32325,32329, 191 32332,32336,32339,32342,32346,32349,32352,32356,32359,32362,32365,32368,32371, 192 32374,32377,32381,32384,32387,32389,32392,32395,32398,32401,32404,32407,32410, 193 32412,32415,32418,32421,32423,32426,32429,32431,32434,32437,32439,32442,32444, 194 32447,32449,32452,32454,32457,32459,32461,32464,32466,32469,32471,32473,32476, 195 32478,32480,32482,32485,32487,32489,32491,32493,32495,32497,32500,32502,32504, 196 32506,32508,32510,32512,32514,32516,32518,32520,32522,32524,32526,32527,32529, 197 32531,32533,32535,32537,32538,32540,32542,32544,32545,32547}; 198 199 static uma_zone_t qdiffsample_zone; 200 201 static MALLOC_DEFINE(M_CDG, "cdg data", 202 "Per connection data required for the CDG congestion control algorithm"); 203 204 static int ertt_id; 205 206 static VNET_DEFINE(uint32_t, cdg_alpha_inc); 207 static VNET_DEFINE(uint32_t, cdg_beta_delay); 208 static VNET_DEFINE(uint32_t, cdg_beta_loss); 209 static VNET_DEFINE(uint32_t, cdg_smoothing_factor); 210 static VNET_DEFINE(uint32_t, cdg_exp_backoff_scale); 211 static VNET_DEFINE(uint32_t, cdg_consec_cong); 212 static VNET_DEFINE(uint32_t, cdg_hold_backoff); 213 #define V_cdg_alpha_inc VNET(cdg_alpha_inc) 214 #define V_cdg_beta_delay VNET(cdg_beta_delay) 215 #define V_cdg_beta_loss VNET(cdg_beta_loss) 216 #define V_cdg_smoothing_factor VNET(cdg_smoothing_factor) 217 #define V_cdg_exp_backoff_scale VNET(cdg_exp_backoff_scale) 218 #define V_cdg_consec_cong VNET(cdg_consec_cong) 219 #define V_cdg_hold_backoff VNET(cdg_hold_backoff) 220 221 /* Function prototypes. */ 222 static int cdg_mod_init(void); 223 static int cdg_mod_destroy(void); 224 static void cdg_conn_init(struct cc_var *ccv); 225 static int cdg_cb_init(struct cc_var *ccv); 226 static void cdg_cb_destroy(struct cc_var *ccv); 227 static void cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type); 228 static void cdg_ack_received(struct cc_var *ccv, uint16_t ack_type); 229 230 struct cc_algo cdg_cc_algo = { 231 .name = "cdg", 232 .mod_init = cdg_mod_init, 233 .ack_received = cdg_ack_received, 234 .cb_destroy = cdg_cb_destroy, 235 .cb_init = cdg_cb_init, 236 .conn_init = cdg_conn_init, 237 .cong_signal = cdg_cong_signal, 238 .mod_destroy = cdg_mod_destroy 239 }; 240 241 /* Vnet created and being initialised. */ 242 static void 243 cdg_init_vnet(const void *unused __unused) 244 { 245 246 V_cdg_alpha_inc = 0; 247 V_cdg_beta_delay = 70; 248 V_cdg_beta_loss = 50; 249 V_cdg_smoothing_factor = 8; 250 V_cdg_exp_backoff_scale = 3; 251 V_cdg_consec_cong = 5; 252 V_cdg_hold_backoff = 5; 253 } 254 255 static int 256 cdg_mod_init(void) 257 { 258 VNET_ITERATOR_DECL(v); 259 260 ertt_id = khelp_get_id("ertt"); 261 if (ertt_id <= 0) 262 return (EINVAL); 263 264 qdiffsample_zone = uma_zcreate("cdg_qdiffsample", 265 sizeof(struct qdiff_sample), NULL, NULL, NULL, NULL, 0, 0); 266 267 VNET_LIST_RLOCK(); 268 VNET_FOREACH(v) { 269 CURVNET_SET(v); 270 cdg_init_vnet(NULL); 271 CURVNET_RESTORE(); 272 } 273 VNET_LIST_RUNLOCK(); 274 275 cdg_cc_algo.post_recovery = newreno_cc_algo.post_recovery; 276 cdg_cc_algo.after_idle = newreno_cc_algo.after_idle; 277 278 return (0); 279 } 280 281 static int 282 cdg_mod_destroy(void) 283 { 284 285 uma_zdestroy(qdiffsample_zone); 286 return (0); 287 } 288 289 static int 290 cdg_cb_init(struct cc_var *ccv) 291 { 292 struct cdg *cdg_data; 293 294 cdg_data = malloc(sizeof(struct cdg), M_CDG, M_NOWAIT); 295 if (cdg_data == NULL) 296 return (ENOMEM); 297 298 cdg_data->shadow_w = 0; 299 cdg_data->max_qtrend = 0; 300 cdg_data->min_qtrend = 0; 301 cdg_data->queue_state = CDG_Q_UNKNOWN; 302 cdg_data->maxrtt_in_rtt = 0; 303 cdg_data->maxrtt_in_prevrtt = 0; 304 cdg_data->minrtt_in_rtt = INT_MAX; 305 cdg_data->minrtt_in_prevrtt = 0; 306 cdg_data->window_incr = 0; 307 cdg_data->rtt_count = 0; 308 cdg_data->consec_cong_cnt = 0; 309 cdg_data->sample_q_size = V_cdg_smoothing_factor; 310 cdg_data->num_samples = 0; 311 STAILQ_INIT(&cdg_data->qdiffmin_q); 312 STAILQ_INIT(&cdg_data->qdiffmax_q); 313 314 ccv->cc_data = cdg_data; 315 316 return (0); 317 } 318 319 static void 320 cdg_conn_init(struct cc_var *ccv) 321 { 322 struct cdg *cdg_data = ccv->cc_data; 323 324 /* 325 * Initialise the shadow_cwnd in case we are competing with loss based 326 * flows from the start 327 */ 328 cdg_data->shadow_w = CCV(ccv, snd_cwnd); 329 } 330 331 static void 332 cdg_cb_destroy(struct cc_var *ccv) 333 { 334 struct cdg *cdg_data; 335 struct qdiff_sample *qds, *qds_n; 336 337 cdg_data = ccv->cc_data; 338 339 qds = STAILQ_FIRST(&cdg_data->qdiffmin_q); 340 while (qds != NULL) { 341 qds_n = STAILQ_NEXT(qds, qdiff_lnk); 342 uma_zfree(qdiffsample_zone,qds); 343 qds = qds_n; 344 } 345 346 qds = STAILQ_FIRST(&cdg_data->qdiffmax_q); 347 while (qds != NULL) { 348 qds_n = STAILQ_NEXT(qds, qdiff_lnk); 349 uma_zfree(qdiffsample_zone,qds); 350 qds = qds_n; 351 } 352 353 free(ccv->cc_data, M_CDG); 354 } 355 356 static int 357 cdg_beta_handler(SYSCTL_HANDLER_ARGS) 358 { 359 360 if (req->newptr != NULL && 361 (CAST_PTR_INT(req->newptr) == 0 || CAST_PTR_INT(req->newptr) > 100)) 362 return (EINVAL); 363 364 return (sysctl_handle_int(oidp, arg1, arg2, req)); 365 } 366 367 static int 368 cdg_exp_backoff_scale_handler(SYSCTL_HANDLER_ARGS) 369 { 370 371 if (req->newptr != NULL && CAST_PTR_INT(req->newptr) < 1) 372 return (EINVAL); 373 374 return (sysctl_handle_int(oidp, arg1, arg2, req)); 375 } 376 377 static inline uint32_t 378 cdg_window_decrease(struct cc_var *ccv, unsigned long owin, unsigned int beta) 379 { 380 381 return ((ulmin(CCV(ccv, snd_wnd), owin) * beta) / 100); 382 } 383 384 /* 385 * Window increase function 386 * This window increase function is independent of the initial window size 387 * to ensure small window flows are not discriminated against (i.e. fairness). 388 * It increases at 1pkt/rtt like Reno for alpha_inc rtts, and then 2pkts/rtt for 389 * the next alpha_inc rtts, etc. 390 */ 391 static void 392 cdg_window_increase(struct cc_var *ccv, int new_measurement) 393 { 394 struct cdg *cdg_data; 395 int incr, s_w_incr; 396 397 cdg_data = ccv->cc_data; 398 incr = s_w_incr = 0; 399 400 if (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)) { 401 /* Slow start. */ 402 incr = CCV(ccv, t_maxseg); 403 s_w_incr = incr; 404 cdg_data->window_incr = cdg_data->rtt_count = 0; 405 } else { 406 /* Congestion avoidance. */ 407 if (new_measurement) { 408 s_w_incr = CCV(ccv, t_maxseg); 409 if (V_cdg_alpha_inc == 0) { 410 incr = CCV(ccv, t_maxseg); 411 } else { 412 if (++cdg_data->rtt_count >= V_cdg_alpha_inc) { 413 cdg_data->window_incr++; 414 cdg_data->rtt_count = 0; 415 } 416 incr = CCV(ccv, t_maxseg) * 417 cdg_data->window_incr; 418 } 419 } 420 } 421 422 if (cdg_data->shadow_w > 0) 423 cdg_data->shadow_w = ulmin(cdg_data->shadow_w + s_w_incr, 424 TCP_MAXWIN << CCV(ccv, snd_scale)); 425 426 CCV(ccv, snd_cwnd) = ulmin(CCV(ccv, snd_cwnd) + incr, 427 TCP_MAXWIN << CCV(ccv, snd_scale)); 428 } 429 430 static void 431 cdg_cong_signal(struct cc_var *ccv, uint32_t signal_type) 432 { 433 struct cdg *cdg_data = ccv->cc_data; 434 uint32_t cwin; 435 u_int mss; 436 437 cwin = CCV(ccv, snd_cwnd); 438 mss = CCV(ccv, t_maxseg); 439 440 switch(signal_type) { 441 case CC_CDG_DELAY: 442 CCV(ccv, snd_ssthresh) = cdg_window_decrease(ccv, 443 CCV(ccv, snd_cwnd), V_cdg_beta_delay); 444 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 445 CCV(ccv, snd_recover) = CCV(ccv, snd_max); 446 cdg_data->window_incr = cdg_data->rtt_count = 0; 447 ENTER_CONGRECOVERY(CCV(ccv, t_flags)); 448 break; 449 case CC_NDUPACK: 450 /* 451 * If already responding to congestion OR we have guessed no 452 * queue in the path is full. 453 */ 454 if (IN_CONGRECOVERY(CCV(ccv, t_flags)) || 455 cdg_data->queue_state < CDG_Q_FULL) { 456 CCV(ccv, snd_ssthresh) = cwin; 457 CCV(ccv, snd_recover) = CCV(ccv, snd_max); 458 } else { 459 /* 460 * Loss is likely to be congestion related. We have 461 * inferred a queue full state, so have shadow window 462 * react to loss as NewReno would. 463 */ 464 if (cdg_data->shadow_w > 0) 465 cdg_data->shadow_w = cdg_window_decrease(ccv, 466 cdg_data->shadow_w, RENO_BETA); 467 468 CCV(ccv, snd_ssthresh) = max(cdg_data->shadow_w, 469 cdg_window_decrease(ccv, cwin, V_cdg_beta_loss)); 470 CCV(ccv, snd_cwnd) = CCV(ccv, snd_ssthresh); 471 472 cdg_data->window_incr = cdg_data->rtt_count = 0; 473 } 474 ENTER_RECOVERY(CCV(ccv, t_flags)); 475 break; 476 case CC_RTO: 477 CCV(ccv, snd_ssthresh) = 478 max((CCV(ccv, snd_max) - CCV(ccv, snd_una)) / 2 / mss, 2) 479 * mss; 480 CCV(ccv, snd_cwnd) = mss; 481 break; 482 default: 483 newreno_cc_algo.cong_signal(ccv, signal_type); 484 break; 485 } 486 } 487 488 /* 489 * Using a negative exponential probabilistic backoff so that sources with 490 * varying RTTs which share the same link will, on average, have the same 491 * probability of backoff over time. 492 * 493 * Prob_backoff = 1 - exp(-qtrend / V_cdg_exp_backoff_scale), where 494 * V_cdg_exp_backoff_scale is the average qtrend for the exponential backoff. 495 */ 496 static inline int 497 prob_backoff(long qtrend) 498 { 499 int backoff, idx, p; 500 501 backoff = (qtrend > ((MAXGRAD * V_cdg_exp_backoff_scale) << D_P_E)); 502 503 if (!backoff) { 504 if (V_cdg_exp_backoff_scale > 1) 505 idx = (qtrend + V_cdg_exp_backoff_scale / 2) / 506 V_cdg_exp_backoff_scale; 507 else 508 idx = qtrend; 509 510 /* Backoff probability proportional to rate of queue growth. */ 511 p = (INT_MAX / (1 << EXP_PREC)) * probexp[idx]; 512 backoff = (random() < p); 513 } 514 515 return (backoff); 516 } 517 518 static inline void 519 calc_moving_average(struct cdg *cdg_data, long qdiff_max, long qdiff_min) 520 { 521 struct qdiff_sample *qds; 522 523 ++cdg_data->num_samples; 524 if (cdg_data->num_samples > cdg_data->sample_q_size) { 525 /* Minimum RTT. */ 526 qds = STAILQ_FIRST(&cdg_data->qdiffmin_q); 527 cdg_data->min_qtrend = cdg_data->min_qtrend + 528 (qdiff_min - qds->qdiff) / cdg_data->sample_q_size; 529 STAILQ_REMOVE_HEAD(&cdg_data->qdiffmin_q, qdiff_lnk); 530 qds->qdiff = qdiff_min; 531 STAILQ_INSERT_TAIL(&cdg_data->qdiffmin_q, qds, qdiff_lnk); 532 533 /* Maximum RTT. */ 534 qds = STAILQ_FIRST(&cdg_data->qdiffmax_q); 535 cdg_data->max_qtrend = cdg_data->max_qtrend + 536 (qdiff_max - qds->qdiff) / cdg_data->sample_q_size; 537 STAILQ_REMOVE_HEAD(&cdg_data->qdiffmax_q, qdiff_lnk); 538 qds->qdiff = qdiff_max; 539 STAILQ_INSERT_TAIL(&cdg_data->qdiffmax_q, qds, qdiff_lnk); 540 --cdg_data->num_samples; 541 } else { 542 qds = uma_zalloc(qdiffsample_zone, M_NOWAIT); 543 if (qds != NULL) { 544 cdg_data->min_qtrend = cdg_data->min_qtrend + 545 qdiff_min / cdg_data->sample_q_size; 546 qds->qdiff = qdiff_min; 547 STAILQ_INSERT_TAIL(&cdg_data->qdiffmin_q, qds, 548 qdiff_lnk); 549 } 550 551 qds = uma_zalloc(qdiffsample_zone, M_NOWAIT); 552 if (qds) { 553 cdg_data->max_qtrend = cdg_data->max_qtrend + 554 qdiff_max / cdg_data->sample_q_size; 555 qds->qdiff = qdiff_max; 556 STAILQ_INSERT_TAIL(&cdg_data->qdiffmax_q, qds, 557 qdiff_lnk); 558 } 559 } 560 } 561 562 static void 563 cdg_ack_received(struct cc_var *ccv, uint16_t ack_type) 564 { 565 struct cdg *cdg_data; 566 struct ertt *e_t; 567 long qdiff_max, qdiff_min; 568 int congestion, new_measurement, slowstart; 569 570 cdg_data = ccv->cc_data; 571 e_t = (struct ertt *)khelp_get_osd(CCV(ccv, osd), ertt_id); 572 new_measurement = e_t->flags & ERTT_NEW_MEASUREMENT; 573 congestion = 0; 574 cdg_data->maxrtt_in_rtt = imax(e_t->rtt, cdg_data->maxrtt_in_rtt); 575 cdg_data->minrtt_in_rtt = imin(e_t->rtt, cdg_data->minrtt_in_rtt); 576 577 if (new_measurement) { 578 slowstart = (CCV(ccv, snd_cwnd) <= CCV(ccv, snd_ssthresh)); 579 /* 580 * Update smoothed gradient measurements. Since we are only 581 * using one measurement per RTT, use max or min rtt_in_rtt. 582 * This is also less noisy than a sample RTT measurement. Max 583 * RTT measurements can have trouble due to OS issues. 584 */ 585 if (cdg_data->maxrtt_in_prevrtt) { 586 qdiff_max = ((long)(cdg_data->maxrtt_in_rtt - 587 cdg_data->maxrtt_in_prevrtt) << D_P_E ); 588 qdiff_min = ((long)(cdg_data->minrtt_in_rtt - 589 cdg_data->minrtt_in_prevrtt) << D_P_E ); 590 591 calc_moving_average(cdg_data, qdiff_max, qdiff_min); 592 593 /* Probabilistic backoff with respect to gradient. */ 594 if (slowstart && qdiff_min > 0) 595 congestion = prob_backoff(qdiff_min); 596 else if (cdg_data->min_qtrend > 0) 597 congestion = prob_backoff(cdg_data->min_qtrend); 598 else if (slowstart && qdiff_max > 0) 599 congestion = prob_backoff(qdiff_max); 600 else if (cdg_data->max_qtrend > 0) 601 congestion = prob_backoff(cdg_data->max_qtrend); 602 603 /* Update estimate of queue state. */ 604 if (cdg_data->min_qtrend > 0 && 605 cdg_data->max_qtrend <= 0) { 606 cdg_data->queue_state = CDG_Q_FULL; 607 } else if (cdg_data->min_qtrend >= 0 && 608 cdg_data->max_qtrend < 0) { 609 cdg_data->queue_state = CDG_Q_EMPTY; 610 cdg_data->shadow_w = 0; 611 } else if (cdg_data->min_qtrend > 0 && 612 cdg_data->max_qtrend > 0) { 613 cdg_data->queue_state = CDG_Q_RISING; 614 } else if (cdg_data->min_qtrend < 0 && 615 cdg_data->max_qtrend < 0) { 616 cdg_data->queue_state = CDG_Q_FALLING; 617 } 618 619 if (cdg_data->min_qtrend < 0 || 620 cdg_data->max_qtrend < 0) 621 cdg_data->consec_cong_cnt = 0; 622 } 623 624 cdg_data->minrtt_in_prevrtt = cdg_data->minrtt_in_rtt; 625 cdg_data->minrtt_in_rtt = INT_MAX; 626 cdg_data->maxrtt_in_prevrtt = cdg_data->maxrtt_in_rtt; 627 cdg_data->maxrtt_in_rtt = 0; 628 e_t->flags &= ~ERTT_NEW_MEASUREMENT; 629 } 630 631 if (congestion) { 632 cdg_data->consec_cong_cnt++; 633 if (!IN_RECOVERY(CCV(ccv, t_flags))) { 634 if (cdg_data->consec_cong_cnt <= V_cdg_consec_cong) 635 cdg_cong_signal(ccv, CC_CDG_DELAY); 636 else 637 /* 638 * We have been backing off but the queue is not 639 * falling. Assume we are competing with 640 * loss-based flows and don't back off for the 641 * next V_cdg_hold_backoff RTT periods. 642 */ 643 if (cdg_data->consec_cong_cnt >= 644 V_cdg_consec_cong + V_cdg_hold_backoff) 645 cdg_data->consec_cong_cnt = 0; 646 647 /* Won't see effect until 2nd RTT. */ 648 cdg_data->maxrtt_in_prevrtt = 0; 649 /* 650 * Resync shadow window in case we are competing with a 651 * loss based flow 652 */ 653 cdg_data->shadow_w = ulmax(CCV(ccv, snd_cwnd), 654 cdg_data->shadow_w); 655 } 656 } else if (ack_type == CC_ACK) 657 cdg_window_increase(ccv, new_measurement); 658 } 659 660 /* When a vnet is created and being initialised, init the per-stack CDG vars. */ 661 VNET_SYSINIT(cdg_init_vnet, SI_SUB_PROTO_BEGIN, SI_ORDER_FIRST, 662 cdg_init_vnet, NULL); 663 664 SYSCTL_DECL(_net_inet_tcp_cc_cdg); 665 SYSCTL_NODE(_net_inet_tcp_cc, OID_AUTO, cdg, CTLFLAG_RW, NULL, 666 "CAIA delay-gradient congestion control related settings"); 667 668 SYSCTL_STRING(_net_inet_tcp_cc_cdg, OID_AUTO, version, 669 CTLFLAG_RD, CDG_VERSION, sizeof(CDG_VERSION) - 1, 670 "Current algorithm/implementation version number"); 671 672 SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, alpha_inc, 673 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_alpha_inc), 0, 674 "Increment the window increase factor alpha by 1 MSS segment every " 675 "alpha_inc RTTs during congestion avoidance mode."); 676 677 SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, beta_delay, 678 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_beta_delay), 70, 679 &cdg_beta_handler, "IU", 680 "Delay-based window decrease factor as a percentage " 681 "(on delay-based backoff, w = w * beta_delay / 100)"); 682 683 SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, beta_loss, 684 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, &VNET_NAME(cdg_beta_loss), 50, 685 &cdg_beta_handler, "IU", 686 "Loss-based window decrease factor as a percentage " 687 "(on loss-based backoff, w = w * beta_loss / 100)"); 688 689 SYSCTL_PROC(_net_inet_tcp_cc_cdg, OID_AUTO, exp_backoff_scale, 690 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 691 &VNET_NAME(cdg_exp_backoff_scale), 2, &cdg_exp_backoff_scale_handler, "IU", 692 "Scaling parameter for the probabilistic exponential backoff"); 693 694 SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, smoothing_factor, 695 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_smoothing_factor), 8, 696 "Number of samples used for moving average smoothing (0 = no smoothing)"); 697 698 SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, loss_compete_consec_cong, 699 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_consec_cong), 5, 700 "Number of consecutive delay-gradient based congestion episodes which will " 701 "trigger loss based CC compatibility"); 702 703 SYSCTL_UINT(_net_inet_tcp_cc_cdg, OID_AUTO, loss_compete_hold_backoff, 704 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(cdg_hold_backoff), 5, 705 "Number of consecutive delay-gradient based congestion episodes to hold " 706 "the window backoff for loss based CC compatibility"); 707 708 DECLARE_CC_MODULE(cdg, &cdg_cc_algo); 709 710 MODULE_DEPEND(cdg, ertt, 1, 1, 1); 711