tcp: cubic fast convergence

Change-Id: I3a15960fe346763faf13e8728ce36c2f3bf7b05a
Signed-off-by: Florin Coras <fcoras@cisco.com>

4 files changed, 73 insertions, 7 deletions

diff --git a/src/vnet/tcp/tcp.c b/src/vnet/tcp/tcp.c
index 2511a176e00..25c2615b101 100644
--- a/src/vnet/tcp/tcp.c
+++ b/src/vnet/tcp/tcp.c
@@ -1448,6 +1448,29 @@ unformat_tcp_cc_algo (unformat_input_t * input, va_list * va)
   return 1;
 }
 
+uword
+unformat_tcp_cc_algo_cfg (unformat_input_t * input, va_list * va)
+{
+  tcp_main_t *tm = vnet_get_tcp_main ();
+  tcp_cc_algorithm_t *cc_alg;
+  unformat_input_t sub_input;
+  int found = 0;
+
+  vec_foreach (cc_alg, tm->cc_algos)
+  {
+    if (!unformat (input, cc_alg->name))
+      continue;
+
+    if (cc_alg->unformat_cfg
+	&& unformat (input, "%U", unformat_vlib_cli_sub_input, &sub_input))
+      {
+	if (cc_alg->unformat_cfg (&sub_input))
+	  found = 1;
+      }
+  }
+  return found;
+}
+
 static clib_error_t *
 tcp_config_fn (vlib_main_t * vm, unformat_input_t * input)
 {
@@ -1472,6 +1495,8 @@ tcp_config_fn (vlib_main_t * vm, unformat_input_t * input)
       else if (unformat (input, "cc-algo %U", unformat_tcp_cc_algo,
 			 &tm->cc_algo))
 	;
+      else if (unformat (input, "%U", unformat_tcp_cc_algo_cfg))
+	;
       else
 	return clib_error_return (0, "unknown input `%U'",
 				  format_unformat_error, input);
diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index d4bebeb47b8..5b235b65d74 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -30,7 +30,7 @@
 #define TCP_PAWS_IDLE 24 * 24 * 60 * 60 * THZ /**< 24 days */
 #define TCP_FIB_RECHECK_PERIOD	1 * THZ	/**< Recheck every 1s */
 #define TCP_MAX_OPTION_SPACE 40
-#define TCP_CC_DATA_SZ 20
+#define TCP_CC_DATA_SZ 24
 
 #define TCP_DUPACK_THRESHOLD 	3
 #define TCP_MAX_RX_FIFO_SIZE 	32 << 20
@@ -336,14 +336,18 @@ typedef struct _tcp_connection
   u32 tx_fifo_size;	/**< Tx fifo size. Used to constrain cwnd */
 } tcp_connection_t;
 
+/* *INDENT-OFF* */
 struct _tcp_cc_algorithm
 {
+  const char *name;
+  uword (*unformat_cfg) (unformat_input_t * input);
   void (*rcv_ack) (tcp_connection_t * tc);
   void (*rcv_cong_ack) (tcp_connection_t * tc, tcp_cc_ack_t ack);
   void (*congestion) (tcp_connection_t * tc);
   void (*recovered) (tcp_connection_t * tc);
   void (*init) (tcp_connection_t * tc);
 };
+/* *INDENT-ON* */
 
 #define tcp_fastrecovery_on(tc) (tc)->flags |= TCP_CONN_FAST_RECOVERY
 #define tcp_fastrecovery_off(tc) (tc)->flags &= ~TCP_CONN_FAST_RECOVERY
@@ -485,7 +489,9 @@ typedef struct _tcp_main
   /** fault-injection */
   f64 buffer_fail_fraction;
 
-  u8 cc_algo;
+  /** Default congestion control algorithm type */
+  tcp_cc_algorithm_type_e cc_algo;
+
 } tcp_main_t;
 
 extern tcp_main_t tcp_main;
diff --git a/src/vnet/tcp/tcp_cubic.c b/src/vnet/tcp/tcp_cubic.c
index b9a1c3da06a..55bbcca3497 100644
--- a/src/vnet/tcp/tcp_cubic.c
+++ b/src/vnet/tcp/tcp_cubic.c
@@ -20,6 +20,15 @@
 #define cubic_c		0.4
 #define west_const 	(3 * (1 - beta_cubic) / (1 + beta_cubic))
 
+typedef struct cubic_cfg_
+{
+  u8 fast_convergence;
+} cubic_cfg_t;
+
+static cubic_cfg_t cubic_cfg = {
+  .fast_convergence = 1,
+};
+
 typedef struct cubic_data_
 {
   /** time period (in seconds) needed to increase the current window
@@ -29,7 +38,7 @@ typedef struct cubic_data_
   /** time (in sec) since the start of current congestion avoidance */
   f64 t_start;
 
-  /** Inflection point of the cubic function */
+  /** Inflection point of the cubic function (in snd_mss segments) */
   u32 w_max;
 
 } __clib_packed cubic_data_t;
@@ -84,8 +93,13 @@ static void
 cubic_congestion (tcp_connection_t * tc)
 {
   cubic_data_t *cd = (cubic_data_t *) tcp_cc_data (tc);
+  u32 w_max;
+
+  w_max = tc->cwnd / tc->snd_mss;
+  if (cubic_cfg.fast_convergence && w_max < cd->w_max)
+    w_max = w_max * ((1.0 + beta_cubic) / 2.0);
 
-  cd->w_max = tc->cwnd / tc->snd_mss;
+  cd->w_max = w_max;
   tc->ssthresh = clib_max (tc->cwnd * beta_cubic, 2 * tc->snd_mss);
 }
 
@@ -120,7 +134,7 @@ cubic_rcv_ack (tcp_connection_t * tc)
   rtt_sec = clib_min (tc->mrtt_us, (f64) tc->srtt * TCP_TICK);
 
   w_cubic = W_cubic (cd, t + rtt_sec) * tc->snd_mss;
-  w_aimd = W_est (cd, t, rtt_sec) * tc->snd_mss;
+  w_aimd = (u64) W_est (cd, t, rtt_sec) * tc->snd_mss;
   if (w_cubic < w_aimd)
     {
       tcp_cwnd_accumulate (tc, tc->cwnd, tc->bytes_acked);
@@ -148,7 +162,7 @@ cubic_rcv_ack (tcp_connection_t * tc)
       else
 	{
 	  /* Practically we can't increment so just inflate threshold */
-	  thresh = 1000 * tc->cwnd;
+	  thresh = 50 * tc->cwnd;
 	}
       tcp_cwnd_accumulate (tc, thresh, tc->bytes_acked);
     }
@@ -165,12 +179,32 @@ cubic_conn_init (tcp_connection_t * tc)
   cd->t_start = cubic_time (tc->c_thread_index);
 }
 
+static uword
+cubic_unformat_config (unformat_input_t * input)
+{
+  if (!input)
+    return 0;
+
+  unformat_skip_white_space (input);
+
+  while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+    {
+      if (unformat (input, "no-fast-convergence"))
+	cubic_cfg.fast_convergence = 0;
+      else
+	return 0;
+    }
+  return 1;
+}
+
 const static tcp_cc_algorithm_t tcp_cubic = {
+  .name = "cubic",
+  .unformat_cfg = cubic_unformat_config,
   .congestion = cubic_congestion,
   .recovered = cubic_recovered,
   .rcv_ack = cubic_rcv_ack,
   .rcv_cong_ack = newreno_rcv_cong_ack,
-  .init = cubic_conn_init
+  .init = cubic_conn_init,
 };
 
 clib_error_t *
diff --git a/src/vnet/tcp/tcp_newreno.c b/src/vnet/tcp/tcp_newreno.c
index 420e47a8ca4..e36416cdfe1 100644
--- a/src/vnet/tcp/tcp_newreno.c
+++ b/src/vnet/tcp/tcp_newreno.c
@@ -79,6 +79,7 @@ newreno_conn_init (tcp_connection_t * tc)
 }
 
 const static tcp_cc_algorithm_t tcp_newreno = {
+  .name = "newreno",
   .congestion = newreno_congestion,
   .recovered = newreno_recovered,
   .rcv_ack = newreno_rcv_ack,