History

tcp: allow at most one TLP probe per flight

Previously TLP may send multiple probes of new data in one flight. This happens when the sender is cwnd limited. After the initial TLP containing new data is sent, the sender receives another ACK that acks partial inflight. It may re-arm another TLP timer to send more, if no further ACK returns before the next TLP timeout (PTO) expires. The sender may send in theory a large amount of TLP until send queue is depleted. This only happens if the sender sees such irregular uncommon ACK pattern. But it is generally undesirable behavior during congestion especially. The original TLP design restrict only one TLP probe per inflight as published in "Reducing Web Latency: the Virtue of Gentle Aggression", SIGCOMM 2013. This patch changes TLP to send at most one probe per inflight. Note that if the sender is app-limited, TLP retransmits old data and did not have this issue.

Yuchung Cheng committed 6 years ago

Commit 76be93f1 parent 17ad73e3 files · +18 −12Browse files

Unified · Split

3 files
  1. include/linux/tcp.h +4 −2
  2. net/ipv4/tcp_input.c +6 −5
  3. net/ipv4/tcp_output.c +8 −5
include/linux/tcp.h+4−2
@@ -220,7 +220,9 @@ struct tcp_sock { · 219 unmodified lines
220220 } rack;
221221 u16 advmss; /* Advertised MSS */
222222 u8 compressed_ack;
223− u8 dup_ack_counter;
223+ u8 dup_ack_counter:2,
224+ tlp_retrans:1, /* TLP is a retransmission */
225+ unused:5;
224226 u32 chrono_start; /* Start time in jiffies of a TCP chrono */
225227 u32 chrono_stat[3]; /* Time in jiffies for chrono_stat stats */
226228 u8 chrono_type:2, /* current chronograph type */
@@ -243,7 +245,7 @@ struct tcp_sock { · 16 unmodified lines
243245 save_syn:1, /* Save headers of SYN packet */
244246 is_cwnd_limited:1,/* forward progress limited by snd_cwnd? */
245247 syn_smc:1; /* SYN includes SMC */
246− u32 tlp_high_seq; /* snd_nxt at the time of TLP retransmit. */
248+ u32 tlp_high_seq; /* snd_nxt at the time of TLP */
247249
248250 u32 tcp_tx_delay; /* delay (in usec) added to TX packets */
249251 u64 tcp_wstamp_ns; /* departure time for next sent data packet */
net/ipv4/tcp_input.c+6−5
@@ -3488,10 +3488,8 @@ static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq) · 3487 unmodified lines
34883488 }
34893489 }
34903490
3491−/* This routine deals with acks during a TLP episode.
3492− * We mark the end of a TLP episode on receiving TLP dupack or when
3493− * ack is after tlp_high_seq.
3494− * Ref: loss detection algorithm in draft-dukkipati-tcpm-tcp-loss-probe.
3491+/* This routine deals with acks during a TLP episode and ends an episode by
3492+ * resetting tlp_high_seq. Ref: TLP algorithm in draft-ietf-tcpm-rack
34953493 */
34963494 static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag)
34973495 {
@@ -3500,7 +3498,10 @@ static void tcp_process_tlp_ack(struct sock *sk, u32 ack, int flag) · 2 unmodified lines
35003498 if (before(ack, tp->tlp_high_seq))
35013499 return;
35023500
3503− if (flag & FLAG_DSACKING_ACK) {
3501+ if (!tp->tlp_retrans) {
3502+ /* TLP of new data has been acknowledged */
3503+ tp->tlp_high_seq = 0;
3504+ } else if (flag & FLAG_DSACKING_ACK) {
35043505 /* This DSACK means original and TLP probe arrived; no loss */
35053506 tp->tlp_high_seq = 0;
35063507 } else if (after(ack, tp->tlp_high_seq)) {
net/ipv4/tcp_output.c+8−5
@@ -2624,6 +2624,11 @@ void tcp_send_loss_probe(struct sock *sk) · 2623 unmodified lines
26242624 int pcount;
26252625 int mss = tcp_current_mss(sk);
26262626
2627+ /* At most one outstanding TLP */
2628+ if (tp->tlp_high_seq)
2629+ goto rearm_timer;
2630+
2631+ tp->tlp_retrans = 0;
26272632 skb = tcp_send_head(sk);
26282633 if (skb && tcp_snd_wnd_test(tp, skb, mss)) {
26292634 pcount = tp->packets_out;
@@ -2641,10 +2646,6 @@ void tcp_send_loss_probe(struct sock *sk) · 11 unmodified lines
26412646 return;
26422647 }
26432648
2644− /* At most one outstanding TLP retransmission. */
2645− if (tp->tlp_high_seq)
2646− goto rearm_timer;
2647−
26482649 if (skb_still_in_host_queue(sk, skb))
26492650 goto rearm_timer;
26502651
@@ -2666,10 +2667,12 @@ void tcp_send_loss_probe(struct sock *sk) · 15 unmodified lines
26662667 if (__tcp_retransmit_skb(sk, skb, 1))
26672668 goto rearm_timer;
26682669
2670+ tp->tlp_retrans = 1;
2671+
2672+probe_sent:
26692673 /* Record snd_nxt for loss detection. */
26702674 tp->tlp_high_seq = tp->snd_nxt;
26712675
2672−probe_sent:
26732676 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPLOSSPROBES);
26742677 /* Reset s.t. tcp_rearm_rto will restart timer from now */
26752678 inet_csk(sk)->icsk_pending = 0;

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t