Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2018 Cisco and/or its affiliates. |
| 3 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | * you may not use this file except in compliance with the License. |
| 5 | * You may obtain a copy of the License at: |
| 6 | * |
| 7 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | * |
| 9 | * Unless required by applicable law or agreed to in writing, software |
| 10 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | * See the License for the specific language governing permissions and |
| 13 | * limitations under the License. |
| 14 | */ |
| 15 | |
| 16 | #include <vnet/tcp/tcp.h> |
| 17 | #include <math.h> |
| 18 | |
| 19 | #define beta_cubic 0.7 |
| 20 | #define cubic_c 0.4 |
| 21 | #define west_const (3 * (1 - beta_cubic) / (1 + beta_cubic)) |
| 22 | |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 23 | typedef struct cubic_cfg_ |
| 24 | { |
| 25 | u8 fast_convergence; |
| 26 | } cubic_cfg_t; |
| 27 | |
| 28 | static cubic_cfg_t cubic_cfg = { |
| 29 | .fast_convergence = 1, |
| 30 | }; |
| 31 | |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 32 | typedef struct cubic_data_ |
| 33 | { |
| 34 | /** time period (in seconds) needed to increase the current window |
| 35 | * size to W_max if there are no further congestion events */ |
| 36 | f64 K; |
| 37 | |
| 38 | /** time (in sec) since the start of current congestion avoidance */ |
| 39 | f64 t_start; |
| 40 | |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 41 | /** Inflection point of the cubic function (in snd_mss segments) */ |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 42 | u32 w_max; |
| 43 | |
| 44 | } __clib_packed cubic_data_t; |
| 45 | |
| 46 | STATIC_ASSERT (sizeof (cubic_data_t) <= TCP_CC_DATA_SZ, "cubic data len"); |
| 47 | |
| 48 | static inline f64 |
| 49 | cubic_time (u32 thread_index) |
| 50 | { |
| 51 | return transport_time_now (thread_index); |
| 52 | } |
| 53 | |
| 54 | /** |
| 55 | * RFC 8312 Eq. 1 |
| 56 | * |
| 57 | * CUBIC window increase function. Time and K need to be provided in seconds. |
| 58 | */ |
| 59 | static inline u64 |
| 60 | W_cubic (cubic_data_t * cd, f64 t) |
| 61 | { |
| 62 | f64 diff = t - cd->K; |
| 63 | |
| 64 | /* W_cubic(t) = C*(t-K)^3 + W_max */ |
| 65 | return cubic_c * diff * diff * diff + cd->w_max; |
| 66 | } |
| 67 | |
| 68 | /** |
| 69 | * RFC 8312 Eq. 2 |
| 70 | */ |
| 71 | static inline f64 |
Florin Coras | 6ebd439 | 2018-11-13 00:03:43 -0800 | [diff] [blame] | 72 | K_cubic (cubic_data_t * cd, u32 wnd) |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 73 | { |
Florin Coras | 6ebd439 | 2018-11-13 00:03:43 -0800 | [diff] [blame] | 74 | /* K = cubic_root(W_max*(1-beta_cubic)/C) |
| 75 | * Because the current window may be less than W_max * beta_cubic because |
| 76 | * of fast convergence, we pass it as parameter */ |
| 77 | return pow ((f64) (cd->w_max - wnd) / cubic_c, 1 / 3.0); |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 78 | } |
| 79 | |
| 80 | /** |
| 81 | * RFC 8312 Eq. 4 |
| 82 | * |
| 83 | * Estimates the window size of AIMD(alpha_aimd, beta_aimd) for |
| 84 | * alpha_aimd=3*(1-beta_cubic)/(1+beta_cubic) and beta_aimd=beta_cubic. |
| 85 | * Time (t) and rtt should be provided in seconds |
| 86 | */ |
| 87 | static inline u32 |
| 88 | W_est (cubic_data_t * cd, f64 t, f64 rtt) |
| 89 | { |
| 90 | /* W_est(t) = W_max*beta_cubic+[3*(1-beta_cubic)/(1+beta_cubic)]*(t/RTT) */ |
| 91 | return cd->w_max * beta_cubic + west_const * (t / rtt); |
| 92 | } |
| 93 | |
| 94 | static void |
| 95 | cubic_congestion (tcp_connection_t * tc) |
| 96 | { |
| 97 | cubic_data_t *cd = (cubic_data_t *) tcp_cc_data (tc); |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 98 | u32 w_max; |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 99 | |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 100 | w_max = tc->cwnd / tc->snd_mss; |
| 101 | if (cubic_cfg.fast_convergence && w_max < cd->w_max) |
| 102 | w_max = w_max * ((1.0 + beta_cubic) / 2.0); |
| 103 | |
| 104 | cd->w_max = w_max; |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 105 | tc->ssthresh = clib_max (tc->cwnd * beta_cubic, 2 * tc->snd_mss); |
| 106 | } |
| 107 | |
| 108 | static void |
| 109 | cubic_recovered (tcp_connection_t * tc) |
| 110 | { |
| 111 | cubic_data_t *cd = (cubic_data_t *) tcp_cc_data (tc); |
| 112 | cd->t_start = cubic_time (tc->c_thread_index); |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 113 | tc->cwnd = tc->ssthresh; |
Florin Coras | 6ebd439 | 2018-11-13 00:03:43 -0800 | [diff] [blame] | 114 | cd->K = K_cubic (cd, tc->cwnd / tc->snd_mss); |
| 115 | } |
| 116 | |
| 117 | static void |
| 118 | cubic_cwnd_accumulate (tcp_connection_t * tc, u32 thresh, u32 bytes_acked) |
| 119 | { |
| 120 | /* We just updated the threshold and don't know how large the previous |
| 121 | * one was. Still, optimistically increase cwnd by one segment and |
| 122 | * clear the accumulated bytes. */ |
| 123 | if (tc->cwnd_acc_bytes > thresh) |
| 124 | { |
| 125 | tc->cwnd += tc->snd_mss; |
| 126 | tc->cwnd_acc_bytes = 0; |
| 127 | } |
| 128 | |
| 129 | tcp_cwnd_accumulate (tc, thresh, tc->bytes_acked); |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 130 | } |
| 131 | |
| 132 | static void |
| 133 | cubic_rcv_ack (tcp_connection_t * tc) |
| 134 | { |
| 135 | cubic_data_t *cd = (cubic_data_t *) tcp_cc_data (tc); |
| 136 | u64 w_cubic, w_aimd; |
| 137 | f64 t, rtt_sec; |
| 138 | u32 thresh; |
| 139 | |
| 140 | /* Constrained by tx fifo, can't grow further */ |
| 141 | if (tc->cwnd >= tc->tx_fifo_size) |
| 142 | return; |
| 143 | |
| 144 | if (tcp_in_slowstart (tc)) |
| 145 | { |
| 146 | tc->cwnd += clib_min (tc->snd_mss, tc->bytes_acked); |
| 147 | return; |
| 148 | } |
| 149 | |
| 150 | t = cubic_time (tc->c_thread_index) - cd->t_start; |
| 151 | rtt_sec = clib_min (tc->mrtt_us, (f64) tc->srtt * TCP_TICK); |
| 152 | |
| 153 | w_cubic = W_cubic (cd, t + rtt_sec) * tc->snd_mss; |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 154 | w_aimd = (u64) W_est (cd, t, rtt_sec) * tc->snd_mss; |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 155 | if (w_cubic < w_aimd) |
| 156 | { |
Florin Coras | 6ebd439 | 2018-11-13 00:03:43 -0800 | [diff] [blame] | 157 | cubic_cwnd_accumulate (tc, tc->cwnd, tc->bytes_acked); |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 158 | } |
| 159 | else |
| 160 | { |
| 161 | if (w_cubic > tc->cwnd) |
| 162 | { |
| 163 | /* For NewReno and slow start, we increment cwnd based on the |
| 164 | * number of bytes acked, not the number of acks received. In |
| 165 | * particular, for NewReno we increment the cwnd by 1 snd_mss |
| 166 | * only after we accumulate 1 cwnd of acked bytes (RFC 3465). |
| 167 | * |
| 168 | * For Cubic, as per RFC 8312 we should increment cwnd by |
| 169 | * (w_cubic - cwnd)/cwnd for each ack. Instead of using that, |
| 170 | * we compute the number of packets that need to be acked |
| 171 | * before adding snd_mss to cwnd and compute the threshold |
| 172 | */ |
| 173 | thresh = (tc->snd_mss * tc->cwnd) / (w_cubic - tc->cwnd); |
| 174 | |
Florin Coras | 6ebd439 | 2018-11-13 00:03:43 -0800 | [diff] [blame] | 175 | /* Make sure we don't increase cwnd more often than every segment */ |
| 176 | thresh = clib_max (thresh, tc->snd_mss); |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 177 | } |
| 178 | else |
| 179 | { |
| 180 | /* Practically we can't increment so just inflate threshold */ |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 181 | thresh = 50 * tc->cwnd; |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 182 | } |
Florin Coras | 6ebd439 | 2018-11-13 00:03:43 -0800 | [diff] [blame] | 183 | cubic_cwnd_accumulate (tc, thresh, tc->bytes_acked); |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 184 | } |
| 185 | } |
| 186 | |
| 187 | static void |
| 188 | cubic_conn_init (tcp_connection_t * tc) |
| 189 | { |
| 190 | cubic_data_t *cd = (cubic_data_t *) tcp_cc_data (tc); |
| 191 | tc->ssthresh = tc->snd_wnd; |
| 192 | tc->cwnd = tcp_initial_cwnd (tc); |
| 193 | cd->w_max = 0; |
| 194 | cd->K = 0; |
| 195 | cd->t_start = cubic_time (tc->c_thread_index); |
| 196 | } |
| 197 | |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 198 | static uword |
| 199 | cubic_unformat_config (unformat_input_t * input) |
| 200 | { |
| 201 | if (!input) |
| 202 | return 0; |
| 203 | |
| 204 | unformat_skip_white_space (input); |
| 205 | |
| 206 | while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) |
| 207 | { |
| 208 | if (unformat (input, "no-fast-convergence")) |
| 209 | cubic_cfg.fast_convergence = 0; |
| 210 | else |
| 211 | return 0; |
| 212 | } |
| 213 | return 1; |
| 214 | } |
| 215 | |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 216 | const static tcp_cc_algorithm_t tcp_cubic = { |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 217 | .name = "cubic", |
| 218 | .unformat_cfg = cubic_unformat_config, |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 219 | .congestion = cubic_congestion, |
| 220 | .recovered = cubic_recovered, |
| 221 | .rcv_ack = cubic_rcv_ack, |
| 222 | .rcv_cong_ack = newreno_rcv_cong_ack, |
Florin Coras | aa01abb | 2018-11-12 09:13:10 -0800 | [diff] [blame] | 223 | .init = cubic_conn_init, |
Florin Coras | 2e31cc3 | 2018-09-25 14:00:34 -0700 | [diff] [blame] | 224 | }; |
| 225 | |
| 226 | clib_error_t * |
| 227 | cubic_init (vlib_main_t * vm) |
| 228 | { |
| 229 | clib_error_t *error = 0; |
| 230 | |
| 231 | tcp_cc_algo_register (TCP_CC_CUBIC, &tcp_cubic); |
| 232 | |
| 233 | return error; |
| 234 | } |
| 235 | |
| 236 | VLIB_INIT_FUNCTION (cubic_init); |
| 237 | |
| 238 | /* |
| 239 | * fd.io coding-style-patch-verification: ON |
| 240 | * |
| 241 | * Local Variables: |
| 242 | * eval: (c-set-style "gnu") |
| 243 | * End: |
| 244 | */ |