tcp: improve rate estimate

Type:feature

- sample rtt estimation
- report acked+sacked
- report last lost bytes
- use snd_una == snd_nxt to detect 0 bytes in flight

Change-Id: I83181261fdb375c7e33d24b7a82343561e6a905f
Signed-off-by: Florin Coras <fcoras@cisco.com>

2 files changed, 22 insertions, 9 deletions

diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index d736685771d..7ccc06aea9c 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -249,18 +249,24 @@ typedef struct tcp_bt_sample_
   u32 prev;			/**< Previous sample index in list */
   u32 min_seq;			/**< Min seq number in sample */
   u32 max_seq;			/**< Max seq number. Set for rxt samples */
-  u64 delivered;		/**< Total delivered when sample taken */
+  u64 delivered;		/**< Total delivered bytes for sample */
   f64 delivered_time;		/**< Delivered time when sample taken */
+  f64 tx_time;			/**< Transmit time for the burst */
   u64 tx_rate;			/**< Tx pacing rate */
   tcp_bts_flags_t flags;	/**< Sample flag */
 } tcp_bt_sample_t;
 
 typedef struct tcp_rate_sample_
 {
-  u64 sample_delivered;		/**< Delivered of sample used for rate */
-  u32 delivered;		/**< Bytes delivered in ack time */
-  f64 ack_time;			/**< Time to ack the bytes delivered */
+  u64 prior_delivered;		/**< Delivered of sample used for rate, i.e.,
+				     total bytes delivered at prior_time */
+  f64 prior_time;		/**< Delivered time of sample used for rate */
+  f64 interval_time;		/**< Time to ack the bytes delivered */
+  f64 rtt_time;			/**< RTT for sample */
   u64 tx_rate;			/**< Tx pacing rate */
+  u32 delivered;		/**< Bytes delivered in interval_time */
+  u32 acked_and_sacked;		/**< Bytes acked + sacked now */
+  u32 lost;			/**< Bytes lost now */
   tcp_bts_flags_t flags;	/**< Rate sample flags from bt sample */
 } tcp_rate_sample_t;
 
diff --git a/src/vnet/tcp/tcp_bt.c b/src/vnet/tcp/tcp_bt.c
index 74947474b53..dd3d9539d9a 100644
--- a/src/vnet/tcp/tcp_bt.c
+++ b/src/vnet/tcp/tcp_bt.c
@@ -242,6 +242,7 @@ tcp_bt_alloc_tx_sample (tcp_connection_t * tc, u32 min_seq)
   bts->delivered = tc->delivered;
   bts->delivered_time = tc->delivered_time;
   bts->tx_rate = transport_connection_tx_pacer_rate (&tc->connection);
+  bts->tx_time = tcp_time_now_us (tc->c_thread_index);
   bts->flags |= tc->app_limited ? TCP_BTS_IS_APP_LIMITED : 0;
   return bts;
 }
@@ -268,7 +269,7 @@ tcp_bt_track_tx (tcp_connection_t * tc)
   tcp_bt_sample_t *bts, *tail;
   u32 bts_index;
 
-  if (!tcp_flight_size (tc))
+  if (tc->snd_una == tc->snd_nxt)
     tc->delivered_time = tcp_time_now_us (tc->c_thread_index);
 
   bts = tcp_bt_alloc_tx_sample (tc, tc->snd_nxt);
@@ -415,12 +416,12 @@ static void
 tcp_bt_sample_to_rate_sample (tcp_connection_t * tc, tcp_bt_sample_t * bts,
 			      tcp_rate_sample_t * rs)
 {
-  if (rs->sample_delivered && rs->sample_delivered >= bts->delivered)
+  if (rs->prior_delivered && rs->prior_delivered >= bts->delivered)
     return;
 
-  rs->sample_delivered = bts->delivered;
-  rs->delivered = tc->delivered - bts->delivered;
-  rs->ack_time = tc->delivered_time - bts->delivered_time;
+  rs->prior_delivered = bts->delivered;
+  rs->prior_time = bts->delivered_time;
+  rs->rtt_time = bts->tx_time;
   rs->tx_rate = bts->tx_rate;
   rs->flags = bts->flags;
 }
@@ -527,6 +528,12 @@ tcp_bt_sample_delivery_rate (tcp_connection_t * tc, tcp_rate_sample_t * rs)
 
   if (tc->sack_sb.last_sacked_bytes)
     tcp_bt_walk_samples_ooo (tc, rs);
+
+  rs->interval_time = tc->delivered_time - rs->prior_time;
+  rs->delivered = tc->delivered - rs->prior_delivered;
+  rs->rtt_time = tc->delivered_time - rs->rtt_time;
+  rs->acked_and_sacked = delivered;
+  rs->lost = tc->sack_sb.last_lost_bytes;
 }
 
 void