Message ID | 1313739212-2315-1-git-send-email-nanditad@google.com |
---|---|
State | Changes Requested, archived |
Delegated to: | David Miller |
Headers | show |
On Fri, 19 Aug 2011, Nandita Dukkipati wrote: > +static void tcp_update_cwnd_in_recovery(struct sock *sk, int newly_acked_sacked, > + int fast_rexmit, int flag) > +{ > + struct tcp_sock *tp = tcp_sk(sk); > + int sndcnt = 0; > + int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); > + > + if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { > + if (WARN_ON(!tp->prior_cwnd)) > + tp->prior_cwnd = 1; This should still be made larger to avoid problems if it ever will be needed. > + sndcnt = DIV_ROUND_UP((u64)(tp->prr_delivered * > + tp->snd_ssthresh), > + (u64)tp->prior_cwnd) - tp->prr_out; I think you should pick one from include/linux/math64.h instead of letting gcc to do / operand all by itself. ...Obviosly then the ROUND_UP part needs to be done manually (either using the remainder or pre-addition like DIV_ROUND_UP does).
From: Nandita Dukkipati <nanditad@google.com> Date: Fri, 19 Aug 2011 00:33:32 -0700 > @@ -2830,9 +2830,14 @@ static int tcp_try_undo_loss(struct sock *sk) > static inline void tcp_complete_cwr(struct sock *sk) > { > struct tcp_sock *tp = tcp_sk(sk); > - /* Do not moderate cwnd if it's already undone in cwr or recovery */ > - if (tp->undo_marker && tp->snd_cwnd > tp->snd_ssthresh) { > - tp->snd_cwnd = tp->snd_ssthresh; > + > + /* Do not moderate cwnd if it's already undone in cwr or recovery. */ > + if (tp->undo_marker) { > + > + if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR) Please get rid of that empty line before the TCP_CA_CWR case. > + sndcnt = DIV_ROUND_UP((u64)(tp->prr_delivered * > + tp->snd_ssthresh), > + (u64)tp->prior_cwnd) - tp->prr_out; This won't link on 32-bit unless __divdi3 libgcc routine is provided by the architecture. To portably do 64-bit division you need to use do_div() or something based upon it. Perhaps DIV_ROUND_UP_LL() will work best in this case. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Forgot to turn off gmail's rich formatting, so re-sending to the list. On Fri, Aug 19, 2011 at 3:25 AM, Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> wrote: > > On Fri, 19 Aug 2011, Nandita Dukkipati wrote: > > > +static void tcp_update_cwnd_in_recovery(struct sock *sk, int newly_acked_sacked, > > + int fast_rexmit, int flag) > > +{ > > + struct tcp_sock *tp = tcp_sk(sk); > > + int sndcnt = 0; > > + int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); > > + > > + if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { > > + if (WARN_ON(!tp->prior_cwnd)) > > + tp->prior_cwnd = 1; > > This should still be made larger to avoid problems if it ever will be > needed. I am letting the value remain at 1, mainly because this is the valid lowest non-zero value for snd_cwnd to take on. The main purpose of this code is to catch any lurking bug outside of PRR which results in an undesirable divide by 0 in PRR. I would like to fix that bug if I find this code is executed. > > > + sndcnt = DIV_ROUND_UP((u64)(tp->prr_delivered * > > + tp->snd_ssthresh), > > + (u64)tp->prior_cwnd) - tp->prr_out; > > I think you should pick one from include/linux/math64.h instead of letting > gcc to do / operand all by itself. ...Obviosly then the ROUND_UP part > needs to be done manually (either using the remainder or pre-addition > like DIV_ROUND_UP does). Done. patch v3 used div_u64. Thanks Nandita -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, Aug 19, 2011 at 3:26 AM, David Miller <davem@davemloft.net> wrote: > From: Nandita Dukkipati <nanditad@google.com> > Date: Fri, 19 Aug 2011 00:33:32 -0700 > >> @@ -2830,9 +2830,14 @@ static int tcp_try_undo_loss(struct sock *sk) >> static inline void tcp_complete_cwr(struct sock *sk) >> { >> struct tcp_sock *tp = tcp_sk(sk); >> - /* Do not moderate cwnd if it's already undone in cwr or recovery */ >> - if (tp->undo_marker && tp->snd_cwnd > tp->snd_ssthresh) { >> - tp->snd_cwnd = tp->snd_ssthresh; >> + >> + /* Do not moderate cwnd if it's already undone in cwr or recovery. */ >> + if (tp->undo_marker) { >> + >> + if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR) > > Please get rid of that empty line before the TCP_CA_CWR case. Done. > >> + sndcnt = DIV_ROUND_UP((u64)(tp->prr_delivered * >> + tp->snd_ssthresh), >> + (u64)tp->prior_cwnd) - tp->prr_out; > > This won't link on 32-bit unless __divdi3 libgcc routine is provided > by the architecture. To portably do 64-bit division you need to use > do_div() or something based upon it. Perhaps DIV_ROUND_UP_LL() will > work best in this case. patch v3 uses div_u64 which is based on do_div() for 32-bit archs. Thanks Nandita -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Fri, 19 Aug 2011, Nandita Dukkipati wrote: > Forgot to turn off gmail's rich formatting, so re-sending to the list. > > On Fri, Aug 19, 2011 at 3:25 AM, Ilpo Järvinen > <ilpo.jarvinen@helsinki.fi> wrote: > > > > On Fri, 19 Aug 2011, Nandita Dukkipati wrote: > > > > > +static void tcp_update_cwnd_in_recovery(struct sock *sk, int newly_acked_sacked, > > > + int fast_rexmit, int flag) > > > +{ > > > + struct tcp_sock *tp = tcp_sk(sk); > > > + int sndcnt = 0; > > > + int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); > > > + > > > + if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { > > > + if (WARN_ON(!tp->prior_cwnd)) > > > + tp->prior_cwnd = 1; > > > > This should still be made larger to avoid problems if it ever will be > > needed. > > I am letting the value remain at 1, mainly because this is the valid > lowest non-zero value for snd_cwnd to take on. The main purpose of > this code is to catch any lurking bug outside of PRR which results in > an undesirable divide by 0 in PRR. I would like to fix that bug if I > find this code is executed. NACK, until this value is at least 2 * tp->snd_ssthresh. Or alternatively the fallback is removed so that we DBZ and do not end up wrecking the network. Other than that I'm ok with the patch (assuming the branches I brought up earlier is ok for everybody else).
On Sat, Aug 20, 2011 at 5:41 AM, Ilpo Järvinen <ilpo.jarvinen@helsinki.fi> wrote: > On Fri, 19 Aug 2011, Nandita Dukkipati wrote: > >> Forgot to turn off gmail's rich formatting, so re-sending to the list. >> >> On Fri, Aug 19, 2011 at 3:25 AM, Ilpo Järvinen >> <ilpo.jarvinen@helsinki.fi> wrote: >> > >> > On Fri, 19 Aug 2011, Nandita Dukkipati wrote: >> > >> > > +static void tcp_update_cwnd_in_recovery(struct sock *sk, int newly_acked_sacked, >> > > + int fast_rexmit, int flag) >> > > +{ >> > > + struct tcp_sock *tp = tcp_sk(sk); >> > > + int sndcnt = 0; >> > > + int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); >> > > + >> > > + if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { >> > > + if (WARN_ON(!tp->prior_cwnd)) >> > > + tp->prior_cwnd = 1; >> > >> > This should still be made larger to avoid problems if it ever will be >> > needed. >> >> I am letting the value remain at 1, mainly because this is the valid >> lowest non-zero value for snd_cwnd to take on. The main purpose of >> this code is to catch any lurking bug outside of PRR which results in >> an undesirable divide by 0 in PRR. I would like to fix that bug if I >> find this code is executed. > > NACK, until this value is at least 2 * tp->snd_ssthresh. Or alternatively > the fallback is removed so that we DBZ and do not end up wrecking the > network. If prior_cwnd is suspect at this point, snd_ssthresh cannot be trusted either. So it appears to me that it's best to remove the fallback and spare the network. Done in patch v4. Nandita > > Other than that I'm ok with the patch (assuming the branches I brought > up earlier is ok for everybody else). > > > -- > i. -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/include/linux/tcp.h b/include/linux/tcp.h index 531ede8..6b63b31 100644 --- a/include/linux/tcp.h +++ b/include/linux/tcp.h @@ -379,6 +379,10 @@ struct tcp_sock { u32 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */ u32 snd_cwnd_used; u32 snd_cwnd_stamp; + u32 prior_cwnd; /* Congestion window at start of Recovery. */ + u32 prr_delivered; /* Number of newly delivered packets to + * receiver in Recovery. */ + u32 prr_out; /* Total number of pkts sent during Recovery. */ u32 rcv_wnd; /* Current receiver window */ u32 write_seq; /* Tail(+1) of data held in tcp send buffer */ diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c index ea0d218..41e5f43 100644 --- a/net/ipv4/tcp_input.c +++ b/net/ipv4/tcp_input.c @@ -2830,9 +2830,14 @@ static int tcp_try_undo_loss(struct sock *sk) static inline void tcp_complete_cwr(struct sock *sk) { struct tcp_sock *tp = tcp_sk(sk); - /* Do not moderate cwnd if it's already undone in cwr or recovery */ - if (tp->undo_marker && tp->snd_cwnd > tp->snd_ssthresh) { - tp->snd_cwnd = tp->snd_ssthresh; + + /* Do not moderate cwnd if it's already undone in cwr or recovery. */ + if (tp->undo_marker) { + + if (inet_csk(sk)->icsk_ca_state == TCP_CA_CWR) + tp->snd_cwnd = min(tp->snd_cwnd, tp->snd_ssthresh); + else /* PRR */ + tp->snd_cwnd = tp->snd_ssthresh; tp->snd_cwnd_stamp = tcp_time_stamp; } tcp_ca_event(sk, CA_EVENT_COMPLETE_CWR); @@ -2950,6 +2955,40 @@ void tcp_simple_retransmit(struct sock *sk) } EXPORT_SYMBOL(tcp_simple_retransmit); +/* This function implements the PRR algorithm, specifcally the PRR-SSRB + * (proportional rate reduction with slow start reduction bound) as described in + * http://www.ietf.org/id/draft-mathis-tcpm-proportional-rate-reduction-01.txt. + * It computes the number of packets to send (sndcnt) based on packets newly + * delivered: + * 1) If the packets in flight is larger than ssthresh, PRR spreads the + * cwnd reductions across a full RTT. + * 2) If packets in flight is lower than ssthresh (such as due to excess + * losses and/or application stalls), do not perform any further cwnd + * reductions, but instead slow start up to ssthresh. + */ +static void tcp_update_cwnd_in_recovery(struct sock *sk, int newly_acked_sacked, + int fast_rexmit, int flag) +{ + struct tcp_sock *tp = tcp_sk(sk); + int sndcnt = 0; + int delta = tp->snd_ssthresh - tcp_packets_in_flight(tp); + + if (tcp_packets_in_flight(tp) > tp->snd_ssthresh) { + if (WARN_ON(!tp->prior_cwnd)) + tp->prior_cwnd = 1; + sndcnt = DIV_ROUND_UP((u64)(tp->prr_delivered * + tp->snd_ssthresh), + (u64)tp->prior_cwnd) - tp->prr_out; + } else { + sndcnt = min_t(int, delta, + max_t(int, tp->prr_delivered - tp->prr_out, + newly_acked_sacked) + 1); + } + + sndcnt = max(sndcnt, (fast_rexmit ? 1 : 0)); + tp->snd_cwnd = tcp_packets_in_flight(tp) + sndcnt; +} + /* Process an event, which can update packets-in-flight not trivially. * Main goal of this function is to calculate new estimate for left_out, * taking into account both packets sitting in receiver's buffer and @@ -2961,7 +3000,8 @@ EXPORT_SYMBOL(tcp_simple_retransmit); * It does _not_ decide what to send, it is made in function * tcp_xmit_retransmit_queue(). */ -static void tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag) +static void tcp_fastretrans_alert(struct sock *sk, int pkts_acked, + int newly_acked_sacked, int flag) { struct inet_connection_sock *icsk = inet_csk(sk); struct tcp_sock *tp = tcp_sk(sk); @@ -3111,13 +3151,17 @@ static void tcp_fastretrans_alert(struct sock *sk, int pkts_acked, int flag) tp->bytes_acked = 0; tp->snd_cwnd_cnt = 0; + tp->prior_cwnd = tp->snd_cwnd; + tp->prr_delivered = 0; + tp->prr_out = 0; tcp_set_ca_state(sk, TCP_CA_Recovery); fast_rexmit = 1; } if (do_lost || (tcp_is_fack(tp) && tcp_head_timedout(sk))) tcp_update_scoreboard(sk, fast_rexmit); - tcp_cwnd_down(sk, flag); + tp->prr_delivered += newly_acked_sacked; + tcp_update_cwnd_in_recovery(sk, newly_acked_sacked, fast_rexmit, flag); tcp_xmit_retransmit_queue(sk); } @@ -3632,6 +3676,8 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag) u32 prior_in_flight; u32 prior_fackets; int prior_packets; + int prior_sacked = tp->sacked_out; + int newly_acked_sacked = 0; int frto_cwnd = 0; /* If the ack is older than previous acks @@ -3703,6 +3749,9 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag) /* See if we can take anything off of the retransmit queue. */ flag |= tcp_clean_rtx_queue(sk, prior_fackets, prior_snd_una); + newly_acked_sacked = (prior_packets - prior_sacked) - + (tp->packets_out - tp->sacked_out); + if (tp->frto_counter) frto_cwnd = tcp_process_frto(sk, flag); /* Guarantee sacktag reordering detection against wrap-arounds */ @@ -3715,7 +3764,7 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag) tcp_may_raise_cwnd(sk, flag)) tcp_cong_avoid(sk, ack, prior_in_flight); tcp_fastretrans_alert(sk, prior_packets - tp->packets_out, - flag); + newly_acked_sacked, flag); } else { if ((flag & FLAG_DATA_ACKED) && !frto_cwnd) tcp_cong_avoid(sk, ack, prior_in_flight); diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 882e0b0..ca50408 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -1796,11 +1796,13 @@ static int tcp_write_xmit(struct sock *sk, unsigned int mss_now, int nonagle, tcp_event_new_data_sent(sk, skb); tcp_minshall_update(tp, mss_now, skb); - sent_pkts++; + sent_pkts += tcp_skb_pcount(skb); if (push_one) break; } + if (inet_csk(sk)->icsk_ca_state == TCP_CA_Recovery) + tp->prr_out += sent_pkts; if (likely(sent_pkts)) { tcp_cwnd_validate(sk); @@ -2294,6 +2296,9 @@ begin_fwd: return; NET_INC_STATS_BH(sock_net(sk), mib_idx); + if (inet_csk(sk)->icsk_ca_state == TCP_CA_Recovery) + tp->prr_out += tcp_skb_pcount(skb); + if (skb == tcp_write_queue_head(sk)) inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, inet_csk(sk)->icsk_rto,
This patch implements Proportional Rate Reduction (PRR) for TCP. PRR is an algorithm that determines TCP's sending rate in fast recovery. PRR avoids excessive window reductions and aims for the actual congestion window size at the end of recovery to be as close as possible to the window determined by the congestion control algorithm. PRR also improves accuracy of the amount of data sent during loss recovery. The patch implements the recommended flavor of PRR called PRR-SSRB (Proportional rate reduction with slow start reduction bound) and replaces the existing rate halving algorithm. PRR improves upon the existing Linux fast recovery under a number of conditions including: 1) burst losses where the losses implicitly reduce the amount of outstanding data (pipe) below the ssthresh value selected by the congestion control algorithm and, 2) losses near the end of short flows where application runs out of data to send. As an example, with the existing rate halving implementation a single loss event can cause a connection carrying short Web transactions to go into the slow start mode after the recovery. This is because during recovery Linux pulls the congestion window down to packets_in_flight+1 on every ACK. A short Web response often runs out of new data to send and its pipe reduces to zero by the end of recovery when all its packets are drained from the network. Subsequent HTTP responses using the same connection will have to slow start to raise cwnd to ssthresh. PRR on the other hand aims for the cwnd to be as close as possible to ssthresh by the end of recovery. A description of PRR and a discussion of its performance can be found at the following links: - IETF Draft: http://tools.ietf.org/html/draft-mathis-tcpm-proportional-rate-reduction-01 - IETF Slides: http://www.ietf.org/proceedings/80/slides/tcpm-6.pdf http://tools.ietf.org/agenda/81/slides/tcpm-2.pdf - Paper to appear in Internet Measurements Conference (IMC) 2011: Improving TCP Loss Recovery Nandita Dukkipati, Matt Mathis, Yuchung Cheng Signed-off-by: Nandita Dukkipati <nanditad@google.com> --- Changelog since v1: - Took care of overflow for large congestion windows in tcp_update_cwnd_in_recovery(). - Renamed prr_cwnd to prior_cwnd. - Renamed pkts_delivered to newly_acked_sacked. include/linux/tcp.h | 4 +++ net/ipv4/tcp_input.c | 61 ++++++++++++++++++++++++++++++++++++++++++++----- net/ipv4/tcp_output.c | 7 +++++- 3 files changed, 65 insertions(+), 7 deletions(-)