summaryrefslogtreecommitdiffstats
path: root/src/vnet/tcp/tcp.h
diff options
context:
space:
mode:
authorFlorin Coras <fcoras@cisco.com>2019-06-12 15:38:19 -0700
committerDave Barach <openvpp@barachs.net>2019-06-25 16:02:51 +0000
commit52814737c351b394d28a8b0ee1544176180f45e0 (patch)
tree76c8f32699624a6beb9554d4effa646c6270a791 /src/vnet/tcp/tcp.h
parent6e5baf29e2e48dc62a439d148f243dbb735de786 (diff)
tcp: delivery rate estimator
Type: feature First cut implementation with limited testing. The feature is not enabled by default and the expectation is that cc algorithms will enable it on demand. Change-Id: I92b70cb4dabcff0e9ccd1d725952c4880af394da Signed-off-by: Florin Coras <fcoras@cisco.com>
Diffstat (limited to 'src/vnet/tcp/tcp.h')
-rw-r--r--src/vnet/tcp/tcp.h139
1 files changed, 127 insertions, 12 deletions
diff --git a/src/vnet/tcp/tcp.h b/src/vnet/tcp/tcp.h
index bc1e3c0f813..b0c3ecc8c15 100644
--- a/src/vnet/tcp/tcp.h
+++ b/src/vnet/tcp/tcp.h
@@ -128,6 +128,8 @@ extern timer_expiration_handler tcp_timer_retransmit_syn_handler;
_(DEQ_PENDING, "Pending dequeue acked") \
_(PSH_PENDING, "PSH pending") \
_(FINRCVD, "FIN received") \
+ _(RATE_SAMPLE, "Conn does rate sampling") \
+ _(TRACK_BURST, "Track burst") \
typedef enum _tcp_connection_flag_bits
{
@@ -174,7 +176,7 @@ typedef struct _sack_scoreboard
u32 tail; /**< Index of last entry */
u32 sacked_bytes; /**< Number of bytes sacked in sb */
u32 last_sacked_bytes; /**< Number of bytes last sacked */
- u32 last_bytes_delivered; /**< Number of sack bytes delivered */
+ u32 last_bytes_delivered; /**< Sack bytes delivered to app */
u32 snd_una_adv; /**< Bytes to add to snd_una */
u32 high_sacked; /**< Highest byte sacked (fack) */
u32 high_rxt; /**< Highest retransmitted sequence */
@@ -231,6 +233,44 @@ void scoreboard_clear (sack_scoreboard_t * sb);
void scoreboard_init (sack_scoreboard_t * sb);
u8 *format_tcp_scoreboard (u8 * s, va_list * args);
+#define TCP_BTS_INVALID_INDEX ((u32)~0)
+
+typedef enum tcp_bts_flags_
+{
+ TCP_BTS_IS_RXT = 1,
+ TCP_BTS_IS_APP_LIMITED = 1 << 1,
+} __clib_packed tcp_bts_flags_t;
+
+typedef struct tcp_bt_sample_
+{
+ u32 next; /**< Next sample index in list */
+ 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 */
+ f64 delivered_time; /**< Delivered time when sample taken */
+ 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 tx_rate; /**< Tx pacing rate */
+ tcp_bts_flags_t flags; /**< Rate sample flags from bt sample */
+} tcp_rate_sample_t;
+
+typedef struct tcp_byte_tracker_
+{
+ tcp_bt_sample_t *samples; /**< Pool of samples */
+ rb_tree_t sample_lookup; /**< Rbtree for sample lookup by min_seq */
+ u32 head; /**< Head of samples linked list */
+ u32 tail; /**< Tail of samples linked list */
+ u32 last_ooo; /**< Cached last ooo sample */
+} tcp_byte_tracker_t;
+
typedef enum _tcp_cc_algorithm_type
{
TCP_CC_NEWRENO,
@@ -304,6 +344,7 @@ typedef struct _tcp_connection
u32 snd_rxt_ts; /**< Timestamp when first packet is retransmitted */
u32 tsecr_last_ack; /**< Timestamp echoed to us in last healthy ACK */
u32 snd_congestion; /**< snd_una_max when congestion is detected */
+ u32 tx_fifo_size; /**< Tx fifo size. Used to constrain cwnd */
tcp_cc_algorithm_t *cc_algo; /**< Congestion control algorithm */
u8 cc_data[TCP_CC_DATA_SZ]; /**< Congestion control algo private data */
@@ -316,15 +357,20 @@ typedef struct _tcp_connection
f64 rtt_ts; /**< Timestamp for tracked ACK */
f64 mrtt_us; /**< High precision mrtt from tracked acks */
- u16 mss; /**< Our max seg size that includes options */
- u32 limited_transmit; /**< snd_nxt when limited transmit starts */
- u32 last_fib_check; /**< Last time we checked fib route for peer */
- u32 sw_if_index; /**< Interface for the connection */
- u32 tx_fifo_size; /**< Tx fifo size. Used to constrain cwnd */
-
u32 psh_seq; /**< Add psh header for seg that includes this */
u32 next_node_index; /**< Can be used to control next node in output */
u32 next_node_opaque; /**< Opaque to pass to next node */
+ u32 limited_transmit; /**< snd_nxt when limited transmit starts */
+ u32 sw_if_index; /**< Interface for the connection */
+
+ /* Delivery rate estimation */
+ u64 delivered; /**< Total bytes delivered to peer */
+ u64 app_limited; /**< Delivered when app-limited detected */
+ f64 delivered_time; /**< Time last bytes were acked */
+ tcp_byte_tracker_t *bt; /**< Tx byte tracker */
+
+ u32 last_fib_check; /**< Last time we checked fib route for peer */
+ u16 mss; /**< Our max seg size that includes options */
} tcp_connection_t;
/* *INDENT-OFF* */
@@ -332,8 +378,9 @@ 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 (*rcv_ack) (tcp_connection_t * tc, tcp_rate_sample_t *rs);
+ void (*rcv_cong_ack) (tcp_connection_t * tc, tcp_cc_ack_t ack,
+ tcp_rate_sample_t *rs);
void (*congestion) (tcp_connection_t * tc);
void (*recovered) (tcp_connection_t * tc);
void (*init) (tcp_connection_t * tc);
@@ -636,6 +683,66 @@ void tcp_program_ack (tcp_worker_ctx_t * wrk, tcp_connection_t * tc);
void tcp_program_dupack (tcp_worker_ctx_t * wrk, tcp_connection_t * tc);
void tcp_send_acks (tcp_worker_ctx_t * wrk);
+/*
+ * Rate estimation
+ */
+
+/**
+ * Byte tracker initialize
+ *
+ * @param tc connection for which the byte tracker should be allocated and
+ * initialized
+ */
+void tcp_bt_init (tcp_connection_t * tc);
+/**
+ * Byte tracker cleanup
+ *
+ * @param tc connection for which the byte tracker should be cleaned up
+ */
+void tcp_bt_cleanup (tcp_connection_t * tc);
+/**
+ * Flush byte tracker samples
+ *
+ * @param tc tcp connection for which samples should be flushed
+ */
+void tcp_bt_flush_samples (tcp_connection_t * tc);
+/**
+ * Track a tcp tx burst
+ *
+ * @param tc tcp connection
+ */
+void tcp_bt_track_tx (tcp_connection_t * tc);
+/**
+ * Track a tcp retransmission
+ *
+ * @param tc tcp connection
+ * @param start start sequence number
+ * @param end end sequence number
+ */
+void tcp_bt_track_rxt (tcp_connection_t * tc, u32 start, u32 end);
+/**
+ * Generate a delivery rate sample from recently acked bytes
+ *
+ * @param tc tcp connection
+ * @param rs resulting rate sample
+ */
+void tcp_bt_sample_delivery_rate (tcp_connection_t * tc,
+ tcp_rate_sample_t * rs);
+/**
+ * Check if sample to be generated is app limited
+ *
+ * @param tc tcp connection
+ */
+void tcp_bt_check_app_limited (tcp_connection_t * tc);
+/**
+ * Check if the byte tracker is in sane state
+ *
+ * Should be used only for testing
+ *
+ * @param bt byte tracker
+ */
+int tcp_bt_is_sane (tcp_byte_tracker_t * bt);
+
always_inline u32
tcp_end_seq (tcp_header_t * th, u32 len)
{
@@ -825,12 +932,19 @@ void tcp_connection_tx_pacer_reset (tcp_connection_t * tc, u32 window,
u32 start_bucket);
always_inline void
-tcp_cc_rcv_ack (tcp_connection_t * tc)
+tcp_cc_rcv_ack (tcp_connection_t * tc, tcp_rate_sample_t * rs)
{
- tc->cc_algo->rcv_ack (tc);
+ tc->cc_algo->rcv_ack (tc, rs);
tc->tsecr_last_ack = tc->rcv_opts.tsecr;
}
+static inline void
+tcp_cc_rcv_cong_ack (tcp_connection_t * tc, tcp_cc_ack_t ack_type,
+ tcp_rate_sample_t * rs)
+{
+ tc->cc_algo->rcv_cong_ack (tc, ack_type, rs);
+}
+
always_inline void
tcp_timer_set (tcp_connection_t * tc, u8 timer_id, u32 interval)
{
@@ -959,7 +1073,8 @@ tcp_cc_data (tcp_connection_t * tc)
return (void *) tc->cc_data;
}
-void newreno_rcv_cong_ack (tcp_connection_t * tc, tcp_cc_ack_t ack_type);
+void newreno_rcv_cong_ack (tcp_connection_t * tc, tcp_cc_ack_t ack_type,
+ tcp_rate_sample_t * rs);
/**
* Push TCP header to buffer