tcp: avoid min-RTT overestimation from delayed ACKs

https://patchwork.ozlabs.org/project/netdev/patch/20180117201101.14137-2-ycheng@google.com/

 

Message ID    20180117201101.14137-2-ycheng@google.com
State    Accepted, archived
Delegated to:    David Miller
Headers    show
Series    tcp: do not use RTT from delayed ACKs for min-RTT | expand
Commit Message
Yuchung ChengJan. 17, 2018, 8:11 p.m. UTC
This patch avoids having TCP sender or congestion control
overestimate the min RTT by orders of magnitude. This happens when
all the samples in the windowed filter are one-packet transfer
like small request and health-check like chit-chat, which is farily
common for applications using persistent connections. This patch
tries to conservatively labels and skip RTT samples obtained from
this type of workload.

Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Soheil Hassas Yeganeh <soheil@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/tcp_input.c | 23 +++++++++++++++++++++--
 1 file changed, 21 insertions(+), 2 deletions(-)
Patch

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index ff71b18d9682..2c6797134553 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -97,6 +97,7 @@  int sysctl_tcp_max_orphans __read_mostly = NR_FILE;  #define FLAG_SACK_RENEGING 0x2000 /* snd_una advanced to a sacked seq */  #define FLAG_UPDATE_TS_RECENT  0x4000 /* tcp_replace_ts_recent() */  #define FLAG_NO_CHALLENGE_ACK  0x8000 /* do not call tcp_send_challenge_ack()  */ +#define FLAG_ACK_MAYBE_DELAYED 0x10000 /* Likely a delayed ACK */     #define FLAG_ACKED     (FLAG_DATA_ACKED|FLAG_SYN_ACKED)  #define FLAG_NOT_DUP       (FLAG_DATA|FLAG_WIN_UPDATE|FLAG_ACKED) @@ -2857,11 +2858,18 @@  static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,     *rexmit = REXMIT_LOST;  }    -static void tcp_update_rtt_min(struct sock *sk, u32 rtt_us) +static void tcp_update_rtt_min(struct sock *sk, u32 rtt_us, const int flag)  {     u32 wlen = sock_net(sk)->ipv4.sysctl_tcp_min_rtt_wlen * HZ;     struct tcp_sock *tp = tcp_sk(sk);    +   if ((flag & FLAG_ACK_MAYBE_DELAYED) && rtt_us > tcp_min_rtt(tp)) { +       /* If the remote keeps returning delayed ACKs, eventually +        * the min filter would pick it up and overestimate the +        * prop. delay when it expires. Skip suspected delayed ACKs. +        */ +       return; +   }     minmax_running_min(&tp->rtt_min, wlen, tcp_jiffies32,                rtt_us ? : jiffies_to_usecs(1));  } @@ -2901,7 +2909,7 @@  static bool tcp_ack_update_rtt(struct sock *sk, const int flag,      * always taken together with ACK, SACK, or TS-opts. Any negative      * values will be skipped with the seq_rtt_us < 0 check above.      */ -   tcp_update_rtt_min(sk, ca_rtt_us); +   tcp_update_rtt_min(sk, ca_rtt_us, flag);     tcp_rtt_estimator(sk, seq_rtt_us);     tcp_set_rto(sk);    @@ -3125,6 +3133,17 @@  static int tcp_clean_rtx_queue(struct sock *sk, u32 prior_fack,     if (likely(first_ackt) && !(flag & FLAG_RETRANS_DATA_ACKED)) {         seq_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, first_ackt);         ca_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, last_ackt); + +       if (pkts_acked == 1 && last_in_flight < tp->mss_cache && +           last_in_flight && !prior_sacked && fully_acked && +           sack->rate->prior_delivered + 1 == tp->delivered && +           !(flag & (FLAG_CA_ALERT | FLAG_SYN_ACKED))) { +           /* Conservatively mark a delayed ACK. It's typically +            * from a lone runt packet over the round trip to +            * a receiver w/o out-of-order or CE events. +            */ +           flag |= FLAG_ACK_MAYBE_DELAYED; +       }     }     if (sack->first_sackt) {         sack_rtt_us = tcp_stamp_us_delta(tp->tcp_mstamp, sack->first_sackt);