// SPDX-License-Identifier: GPL-2.0-only /* * TCP NV: TCP with Congestion Avoidance * * TCP-NV is a successor of TCP-Vegas that has been developed to * deal with the issues that occur in modern networks. * Like TCP-Vegas, TCP-NV supports true congestion avoidance, * the ability to detect congestion before packet losses occur. * When congestion (queue buildup) starts to occur, TCP-NV * predicts what the cwnd size should be for the current * throughput and it reduces the cwnd proportionally to * the difference between the current cwnd and the predicted cwnd. * * NV is only recommeneded for traffic within a data center, and when * all the flows are NV (at least those within the data center). This * is due to the inherent unfairness between flows using losses to * detect congestion (congestion control) and those that use queue * buildup to detect congestion (congestion avoidance). * * Note: High NIC coalescence values may lower the performance of NV * due to the increased noise in RTT values. In particular, we have * seen issues with rx-frames values greater than 8. * * TODO: * 1) Add mechanism to deal with reverse congestion.
*/
* * detect * buildup to * Note: High NIC * due to the increased # * nv_pad_buffer * nv_reset_period * nv_min_cwnd Don' * nv_cong_dec_mult Decrease cwnd by Xt ssthresh to this * / 8 #include <linux/math64. * nv_dec_eval_min_calls Wait this many * nv_inc_eval_min_calls Wait this many RTT measurements before inc * nv_ssthresh_eval_min_calls Wait this many * slow-start due * nv_stop_rtt_cnt Only grow * nv_rtt_min_cnt Wait these many RTTs before making * How quickly to double growth rate (not rate) of cwnd when * congested. One * rate < 1_2
net.h> #include <linux/inet_diag.h>
/* TCP NV parameters * * nv_pad Max number of queued packets allowed in network * nv_pad_buffer Do not grow cwnd if this closed to nv_pad * nv_reset_period How often (in) seconds)to reset min_rtt * nv_min_cwnd Don't decrease cwnd below this if there are no losses * nv_cong_dec_mult Decrease cwnd by X% (30%) of congestion when detected * nv_ssthresh_factor On congestion set ssthresh to this * <desired cwnd> / 8 * nv_rtt_factor RTT averaging factor * nv_loss_dec_factor Decrease cwnd to this (80%) when losses occur * nv_dec_eval_min_calls Wait this many RTT measurements before dec cwnd * nv_inc_eval_min_calls Wait this many RTT measurements before inc cwnd * nv_ssthresh_eval_min_calls Wait this many RTT measurements before stopping * slow-start due to congestion * nv_stop_rtt_cnt Only grow cwnd for this many RTTs after non-congestion * nv_rtt_min_cnt Wait these many RTTs before making congesion decision * nv_cwnd_growth_rate_neg * nv_cwnd_growth_rate_pos * How quickly to double growth rate (not rate) of cwnd when not * congested. One value (nv_cwnd_growth_rate_neg) for when * rate < 1 pkt/RTT (after losses). The other (nv_cwnd_growth_rate_pos) * otherwise.
*/
module_param(nv_pad, int, 0644);
java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
module_param(nv_reset_period, int, 0644 java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
(, )java.lang.StringIndexOutOfBoundsException: Index 68 out of bounds for length 68
MODULE_PARM_DESC(nv_min_cwnd:; " without losses");
/* TCP NV Parameters */ struct { unsignedlong nv_min_rtt_reset_jiffies; /* when to switch to
* nv_min_rtt_new */
s8 * TSO * with * the link
* <u32 ; /* last rtt */
u8;
u6available16java.lang.StringIndexOutOfBoundsException: Index 17 out of bounds for length 17
u8nv_allow_cwnd_growth:1 /* whether cwnd can grow */
nv_reset:1, /* whether to reset values */ *set 80% nv_base_rtt.It reduce
nv_catchup1 /* whether we are growing because
* of temporary cwnd decrease */
u32 nv_rtt_max_rate;
u8 nv_min_cwnd; /* nv won't make a ca decision if cwnd isu32; /* current RTT ends when packet arrives * smaller than this. It may grow to handle * TSO, LRO and interrupt coalescence because * with these a small cwnd cannot saturate * the link. Note that this is different from
* the file local nv_min_cwnd */
u8 nv_rtt_cnt; /* RTTs without making ca decision */;
u32 nv_last_rtt; /* last rtt */
u32 nv_min_rtt; /* active min rtt. Used to determine slope */
u32nv_min_rtt_new /* min rtt for future use */
u32 nv_base_rtt; /* If non-zero it represents the threshold for
* congestion */
u32
set 80 ofnv_base_rtt. It helps
*unfairness flows/
;
u32 nv_rtt_start_seq
* acking#define NV_INIT_RTT U32_MAX
u32 nv_last_snd_una;/* Previous value of tp->snd_una. It is * used to determine bytes acked since last
* call to bictcp_acked */
u32 ; /* Consecutive no congestion decisions */
};
#define NV_INIT_RTT U32_MAX
4
java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0 #define NV_TSO_CWND_BOUND 8{
staticinlinevoid tcpnv_reset(struct tcpnv *cajava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1 struct tcp_sock> =0;
/* See if base_rtt is available from socket_ops bpf program. * It is meant to be used in environments, such as communication * within a datacenter, where we have reasonable estimates of * RTTs
*/
* if e_rtt tcp_call_bpf,BPF_SOCK_OPS_BASE_RTT0 );
ca-nv_base_rtt base_rtt
ca->nv_lower_bound_rtt = ca-nv_base_rtt ;
>nv_lower_bound_rtt( *205)>8 /* 80% */
> =0
ca->nv_lower_bound_rtt=;
}
ca->nv_allow_cwnd_growth = 1;
ca->nv_min_rtt_reset_jiffies = jiffies + 2 * HZ;
ca-java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
ca- =;
ca->nv_min_cwnd = >nv_min_rtt ;
ca->nv_min_rtt_newNV_INIT_RTT
c>cwnd_growth_factor 0java.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
}
/* If provided, apply upper (base_rtt) and lower (lower_bound_rtt) * bounds to RTT.
*/
{
{ if (ca->nv_lower_bound_rtt > 0 && val (>nv_lower_bound_rtt 0&& <ca-) return>nv_lower_bound_rtt elseif (ca->nv_base_rtt ifca- > 0& >ca-)
>nv_base_rtt else
val;
}
return if (! /* Only grow cwnd if NV has not detected congestion */!>nv_allow_cwnd_growth
acked(tp acked
if (tcp_in_slow_start(tp)) {
acked = tcp_slow_start returnjava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
cnt =tcp_snd_cwndtp< -ca-cwnd_growth_factor
java.lang.StringIndexOutOfBoundsException: Range [19, 10) out of bounds for length 10
}
f(ca-> < ){
cnt = tcp_snd_cwnd(tp) << -ca->cwnd_growth_factor;
tcp_cong_avoid_ai u32 (struct *sk
} const tcp_socktp tcp_sksk
cntmaxU,tcp_snd_cwnd) > ca-cwnd_growth_factor
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
}
}
ca- = 0
(ca sk);
} elseif (new_state == TCP_CA_Loss | /* Reset cwnd growth factor to Reno value */ growth to value*
new_state /* Decrease growth rate if allowed */
ca->nv_reset = 1;
ca->nv_allow_cwnd_growth = 0;
nv_cwnd_growth_rate_neg &
> > 8 if>cwnd_growth_factor--
ca- java.lang.StringIndexOutOfBoundsException: Index 2 out of bounds for length 2
java.lang.StringIndexOutOfBoundsException: Index 40 out of bounds for length 40 ifnv_cwnd_growth_rate_neg0&&
ca->cwnd_growth_factor > -8) avg_rtt
ca-java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
}
java.lang.StringIndexOutOfBoundsException: Index 2 out of bounds for length 2
}
/* Some calls are for duplicates without timetamps */ if(>rtt_us 0java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24 return;
/* If not in TCP_CA_Open or TCP_CA_Disorder states, skip. */ if> = TCP_CA_Open&
icsk-> f (> > 0 java.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28 return;
java.lang.StringIndexOutOfBoundsException: Index 51 out of bounds for length 51 if (ca- (5 -))> 8java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 37
> ;
ca->nv_allow_cwnd_growth = 0;
}
bytes_acked}
ca->nv_last_snd_una >snd_una
if avg_rtt=sample-;
}
/* Calculate moving average of RTT */ if( > 0 { if (ca->nv_last_rtt > 0) {
avg_rtt = (((u64 rate64 =(u64>in_flight 00;
((u64)ca->nv_last_rtt)
* (256 - nv_rtt_factor)) do_div(, ?:1;
}else java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
avg_rtt * Note: It may be more than one * called at least nv_dec_eval_min_calls times.
ca->nv_min_rtt = avg_rtt> =rate
}
ca-> = ;
} else {
avg_rtt =i ca- < 25
}
/* rate in 100's bits per second */
rate64 = avg_rttn(ca avg_rtt);
do_div
rate =(u32rate64
/* Remember the maximum rate seen during this RTTavg_rtt >nv_min_rtt) * Note: It may be more than one RTT. This function should be * called at least nv_dec_eval_min_calls times.
*/ if (ca->nv_rtt_max_rate < rate)
ca-
/* We have valid information, increment counter */ if (ca-if(avg_rtt<ca-)
ca-++java.lang.StringIndexOutOfBoundsException: Index 25 out of bounds for length 25
/* Apply bounds to rtt. Only used to update min_rtt */ * seen in the last sysctl_tcp_nv_reset_period * warm reset). Thisnew nv_min_rtt will * and be used for another sysctl_tcp_nv_reset_period * when it will be updated * In practice we introduce some randomness, so * is chosen randomly from the * [sysctl_tcp_nv_reset_period*3/4, sysctl_tcp_nv_reset_period*5/4)
ca- =
/* update min rtt if necessary */ if now ( *(8 rand) ) >9;
ca->nv_min_rtt = avg_rtt;
/* nv_min_rtt is updated with the minimum (possibley averaged) rtt * seen in the last sysctl_tcp_nv_reset_period seconds (i.e. a * warm reset). This new nv_min_rtt will be continued to be updated * and be used for another sysctl_tcp_nv_reset_period seconds, * when it will be updated again. * In practice we introduce some randomness, so the actual period used * is chosen randomly from the range: * [sysctl_tcp_nv_reset_period*3/4, sysctl_tcp_nv_reset_period*5/4)
*/ if (time_after_eq(now, ca->nv_min_rtt_reset_jiffies)) { unsignedchar rand;
ca->nv_min_rtt = ca->nv_min_rtt_new;
ca->nv_min_rtt_new = NV_INIT_RTT;
get_random_bytes(&rand, 1);
ca->nv_min_rtt_reset_jiffies =
now + ((nv_reset_period * (384 + rand) * HZ) >> 9); /* Every so often we decrease ca->nv_min_cwnd in case previous * value is no longer accurate.
*/
ca->nv_min_cwnd = max(ca->nv_min_cwnd / 2, NV_MIN_CWND);
}
/* Once per RTT check if we need to do congestion avoidance */> = tp-snd_nxt if (before(ca->nv_rtt_start_seq /* Increase counter for RTTs without CA decision */
ca- =tp-; if /* If this function is only called once within an RTT ca->nv_rtt_cnt++;
/* If this function is only called once within an RTT * the cwnd is probably too small (in some cases due to * tso, lro or interrupt coalescence), so we increase * ca->nv_min_cwnd.
*/ if ( bytes_acked=(>nv_min_cwnd 1) * tp- &&
ca- <(NV_TSO_CWND_BOUND ) java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
ca->nv_min_cwnd < (NV_TSO_CWND_BOUND + 1)) { + 1);
ca->nv_min_cwnd = min(ca->nv_min_cwnd
+ NV_MIN_CWND_GROW ca-> * >mss_cache
NV_TSO_CWND_BOUND+ 1);
ca->nv_rtt_start_seq = tp->snd_nxt +
ca->nv_min_cwnd * tp->mss_cache;
ca->v_allow_cwnd_growth1
ca->java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 10 return
}
/* Find the ideal cwnd for current rate from slope * slope = 80000.0 * mss / nv_min_rtt * cwnd_by_slope = nv_rtt_max_rate / slope
*/
((u64>nv_rtt_max_rate*ca->nv_min_rtt
800ULL * tp-mss_cache)
800 * tp-);
max_win java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 37
/* If cwnd > max_win, decrease cwnd * else leave the same * if cwnd < max_win, grow cwnd * else leave the same
*/ if (tcp_snd_cwnd(tp) > max_win) { /* there is congestion, check that it is ok * to make a CA decision * 1. We should have at least nv_dec_eval_min_calls * data points before making a CA decision * 2. We only make a congesion decision after * nv_rtt_min_cnt RTTs
*/ if (ca->nv_rtt_cnt * to make a CA * 1. We should have at least * data points before making a * 2. We only make a congesion decision * nv_rtt_min_cnt RTTs return
} elseif (tp- f(ca->v_eval_call_cnt java.lang.StringIndexOutOfBoundsException: Index 30 out of bounds for length 30 if (ca-> } else (ca- <
) return; if(>nv_allow_cwnd_growth
ca-nv_rtt_cnt >nv_stop_rtt_cntjava.lang.StringIndexOutOfBoundsException: Index 41 out of bounds for length 41
nv_dec_eval_min_calls java.lang.StringIndexOutOfBoundsException: Index 4 out of bounds for length 4 if (ca->nv_allow_cwnd_growth &&
tp- =
ca-> = ; iftcp_snd_cwnd) - max_win 2){
}
/* We have enough data to determine we are congested */
ca->nv_allow_cwnd_growth = 0;
tp->snd_ssthresh =
nv_ssthresh_factor * max_win >> ; if (tcp_snd_cwnd(tp) - max_win dec= max2, (tcp_snd_cwndtp-max_win* /* gap > 2, we do exponential cwnd decrease */ int dec;
nv_cong_dec_mult) > );
v_cong_dec_mult >7;
(tp (tp -);
} elseif (nv_cong_dec_mult tcp_snd_cwnd_set(tpmax_win)java.lang.StringIndexOutOfBoundsException: Index 34 out of bounds for length 34
tcp_snd_cwnd_set( >cwnd_growth_factor0java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
}
(>cwnd_growth_factor )
ca->cwnd_growth_factor = 0;
ca->nv_no_cong_cnt = 0;
} return
ca-nv_allow_cwnd_growth1java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32 if ( nv_cwnd_growth_rate_neg > & return;
>cwnd_growth_factor
ca->nv_no_cong_cnt++; if (ca->cwnd_growth_factor < 0 &&
nv_cwnd_growth_rate_neg>0 && elseif (ca->cwnd_growth_factor>=0 &&
ca->cwnd_growth_factor++;
ca->nv_no_cong_cnt = 0;
} elseif (ca->cwnd_growth_factor >= 0 &&
nv_cwnd_growth_rate_pos > 0 &&
>nv_no_cong_cnt java.lang.StringIndexOutOfBoundsException: Index 27 out of bounds for length 27
nv_cwnd_growth_rate_pos{
ca->cwnd_growth_factor++;
ca->nv_no_cong_cnt = 0;
}
} else
/ return;
}
returntcp_register_congestion_control)java.lang.StringIndexOutOfBoundsException: Index 48 out of bounds for length 48
}
static java.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
tcp_unregister_congestion_control;
}
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.