A scenario of tcp performance killer

574 views
Skip to first unread message

mingkun bian

unread,
May 7, 2021, 3:51:24 AM5/7/21
to BBR Development
Hi,

    I found that some tcp packets of http/1.1 are sent very slowly in linux version 4.18.0, then I use bcc to collect some tcp stack and bbr info in tcp_ack、tcp_v4_send_check and tcp_retransmit_timer, then I find some surprising packet as following:
1. In line 66, why did not continue to send the remaining lost packets marked by rack but using rto?(we have changing the max rto to 1 second)
2. After first rto, why did ack not to trigger a new send but using tcp_retransmit_timer?

      GET/ACK _time(us)|seq |ack  |size |snd_nxt |rwnd |cwnd|sndbuf_pacing      |min_rtt_us|rtt_cnt|round_start|full_bw|max_full_bw|write_seq|snd_una|packets_out|sacked_out|lost_out|retrans_out|tcp_state|bbr_state|srtt_us/8
      1 GET     _0       _0       _0       _822     _0       _83968   _47  _0     _9169873     _19745   _0   _0_0       _0       _0       _0  _0 _0 _0 _0 _0 _19745
      2 SND     _10      _0       _822     _0       _0       _83968   _47  _0     _9169873     _19745   _0   _0_0       _0       _0       _0  _0 _0 _0 _0 _0 _19745
      3 SND     _3170    _0       _822     _5313    _0       _83968   _47  _5313  _9169873     _19745   _0   _0_0       _0       _0       _0  _0 _0 _0 _0 _0 _19745
      4 ACK     _27370   _822     _5313    _0       _5313    _83968   _47  _5313  _9169873     _19745   _0   _0_0       _0       _0       _4  _0 _0 _0 _0 _0 _19745
      5 GET     _0       _0       _0       _904     _0       _94208   _51  _0     _9169873     _19745   _1   _1_0       _2773    _0       _0  _0 _0 _0 _0 _0 _20301
      6 SND     _5739    _0       _904     _8088    _0       _94208   _51  _45832 _9169873     _19745   _1   _1_0       _2773    _0       _0  _0 _0 _0 _0 _0 _20301
      7 SND     _6632    _8088    _904     _8088    _8088    _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _6  _0 _0 _0 _0 _0 _20301
      8 SND     _7527    _16176   _904     _8088    _16176   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _12 _0 _0 _0 _0 _0 _20301
      9 SND     _8421    _24264   _904     _8088    _24264   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _18 _0 _0 _0 _0 _0 _20301
     10 SND     _9314    _32352   _904     _8088    _32352   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _24 _0 _0 _0 _0 _0 _20301
     11 SND     _10207   _40440   _904     _8088    _40440   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _30 _0 _0 _0 _0 _0 _20301
     12 SND     _11101   _48528   _904     _5392    _48528   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _36 _0 _0 _0 _0 _0 _20301
     13 SND     _11700   _53920   _904     _8088    _53920   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _40 _0 _0 _0 _0 _0 _20301
     14 SND     _12592   _62008   _904     _6740    _62008   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _46 _0 _0 _0 _0 _0 _20301
     15 ACK     _23791   _904     _1348    _0       _68748   _94208   _51  _230252_9169873     _19745   _1   _1_0       _2773    _0       _51 _0 _0 _0 _0 _0 _20301
     16 SND     _23799   _68748   _904     _2696    _68748   _93184   _52  _228904_9169873     _18053   _2   _1_0       _2773    _1348    _50 _0 _0 _0 _0 _0 _20020
     17 ACK     _33807   _904     _8088    _0       _71444   _93184   _52  _228904_9169873     _18053   _2   _1_0       _2773    _1348    _52 _0 _0 _0 _0 _0 _20020
     18 SND     _33817   _71444   _904     _6740    _71444   _86528   _52  _222164_9169873     _18053   _2   _0_0       _3586    _8088    _47 _0 _0 _0 _0 _0 _21026
     19 ACK     _33827   _904     _14828   _0       _78184   _86528   _52  _222164_9169873     _18053   _2   _0_0       _3586    _8088    _52 _0 _0 _0 _0 _0 _21026
     20 SND     _34583   _78184   _904     _6740    _78184   _79872   _52  _215424_9169873     _18053   _2   _0_0       _6570    _14828   _47 _0 _0 _0 _0 _0 _21797
     21 ACK     _59762   _904     _24264   _0       _84924   _79872   _52  _215424_9169873     _18053   _2   _0_0       _6570    _14828   _52 _0 _0 _0 _0 _0 _21797
     22 SND     _59773   _84924   _904     _8088    _84924   _70656   _52  _205988_9169873     _18053   _2   _0_0       _6570    _24264   _45 _0 _0 _0 _0 _0 _25714
     23 SND     _60672   _93012   _904     _1348    _93012   _70656   _52  _205988_9169873     _18053   _2   _0_0       _6570    _24264   _51 _0 _0 _0 _0 _0 _25714
     24 ACK     _64780   _904     _26960   _0       _94360   _70656   _52  _205988_9169873     _18053   _2   _0_0       _6570    _24264   _52 _0 _0 _0 _0 _0 _25714
     25 ACK     _64789   _904     _32352   _0       _94360   _68096   _52  _203292_9169873     _18053   _2   _0_0       _6570    _26960   _50 _0 _0 _0 _0 _0 _29545
     26 ACK     _68764   _904     _40440   _0       _94360   _62976   _52  _197900_9169873     _18053   _2   _0_0       _6818    _32352   _46 _0 _0 _0 _0 _0 _32898
     27 SND     _68773   _94360   _904     _1376    _94360   _55296   _52  _189812_9169873     _18053   _2   _0_0       _7985    _40440   _40 _0 _0 _0 _0 _0 _36217
     28 ACK     _68782   _904     _43136   _0       _95736   _55296   _52  _189812_9169873     _18053   _2   _0_0       _7985    _40440   _42 _0 _0 _0 _0 _0 _36217
     29 ACK     _68789   _904     _48528   _0       _95736   _52736   _52  _187116_9169873     _18053   _2   _0_0       _8515    _43136   _40 _0 _0 _0 _0 _0 _39012
     30 ACK     _68805   _904     _68748   _0       _95736   _47616   _52  _181724_9169873     _18053   _2   _0_0       _9579    _48528   _36 _0 _0 _0 _0 _0 _41458
     31 ACK     _68813   _904     _83576   _0       _95736   _65536   _67  _161504_9169873     _18053   _2   _0_0       _13567   _68748   _21 _0 _0 _0 _0 _0 _43489
     32 SND     _68933   _95736   _904     _4016    _95736   _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _10 _0 _0 _0 _0 _0 _43679
     33 SND     _69381   _99752   _904     _8088    _99752   _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _13 _0 _0 _0 _0 _0 _43679
     34 SND     _70274   _107840  _904     _8088    _107840  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _19 _0 _0 _0 _0 _0 _43679
     35 SND     _71167   _115928  _904     _8088    _115928  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _25 _0 _0 _0 _0 _0 _43679
     36 SND     _72061   _124016  _904     _8088    _124016  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _31 _0 _0 _0 _0 _0 _43679
     37 SND     _72955   _132104  _904     _8088    _132104  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _37 _0 _0 _0 _0 _0 _43679
     38 SND     _73848   _140192  _904     _5392    _140192  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _43 _0 _0 _0 _0 _0 _43679
     39 SND     _74448   _145584  _904     _3528    _145584  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _47 _0 _0 _0 _0 _0 _43679
     40 ACK     _81885   _904     _86272   _0       _149112  _65536   _78  _146676_9169873     _18053   _3   _1_24456   _24456   _83576   _50 _0 _0 _0 _0 _0 _43679
     41 SND     _81894   _149112  _904     _2696    _149112  _65536   _80  _143980_9169873     _18053   _3   _0_24456   _24456   _86272   _48 _0 _0 _0 _0 _0 _44132
     42 ACK     _98831   _904     _115928  _0       _151808  _65536   _80  _143980_9169873     _18053   _3   _0_24456   _24456   _86272   _50 _0 _0 _0 _0 _0 _44132
     43 SND     _98844   _151808  _904     _8088    _151808  _65536   _103 _114324_9169873     _18053   _4   _1_24456   _24456   _115928  _27 _0 _0 _0 _0 _0 _43498
     44 SND     _99738   _159896  _904     _8088    _159896  _65536   _103 _114324_9169873     _18053   _4   _1_24456   _24456   _115928  _33 _0 _0 _0 _0 _0 _43498
     45 SND     _100631  _167984  _904     _8088    _167984  _65536   _103 _114324_9169873     _18053   _4   _1_24456   _24456   _115928  _39 _0 _0 _0 _0 _0 _43498
     46 SND     _101525  _176072  _904     _5392    _176072  _65536   _103 _114324_9169873     _18053   _4   _1_24456   _24456   _115928  _45 _0 _0 _0 _0 _0 _43498
     47 ACK     _106852  _904     _132104  _0       _181464  _65536   _103 _114324_9169873     _18053   _4   _1_24456   _24456   _115928  _49 _0 _0 _0 _0 _0 _43498
     48 SND     _106861  _181464  _904     _8088    _181464  _65536   _103 _98148 _9169873     _18053   _4   _0_24456   _24456   _132104  _37 _0 _0 _0 _0 _0 _42521
     49 SND     _107753  _189552  _904     _1864    _189552  _65536   _103 _98148 _9169873     _18053   _4   _0_24456   _24456   _132104  _43 _0 _0 _0 _0 _0 _42521
     50 ACK     _111779  _904     _146932  _0       _191416  _65536   _103 _98148 _9169873     _18053   _4   _0_24456   _24456   _132104  _45 _0 _0 _0 _0 _0 _42521
     51 SND     _111788  _191416  _904     _8088    _191416  _65536   _103 _83320 _9169873     _18053   _4   _0_24456   _24456   _146932  _34 _0 _0 _0 _0 _0 _42059
     52 SND     _112681  _199504  _904     _8088    _199504  _65536   _103 _83320 _9169873     _18053   _4   _0_24456   _24456   _146932  _40 _0 _0 _0 _0 _0 _42059
     53 SND     _113576  _207592  _904     _4876    _207592  _65536   _103 _83320 _9169873     _18053   _4   _0_24456   _24456   _146932  _46 _0 _0 _0 _0 _0 _42059
     54 ACK     _116768  _904     _151808  _0       _212468  _65536   _103 _83320 _9169873     _18053   _4   _0_24456   _24456   _146932  _50 _0 _0 _0 _0 _0 _42059
     55 SND     _116776  _212468  _904     _4876    _212468  _65536   _103 _78444 _9169873     _18053   _4   _0_24456   _24456   _151808  _46 _0 _0 _0 _0 _0 _42092
     56 ACK     _131762  _904     _178768  _0       _217344  _65536   _103 _78444 _9169873     _18053   _4   _0_24456   _24456   _151808  _50 _0 _0 _0 _0 _0 _42092
     57 SND     _131773  _217344  _904     _8088    _217344  _65536   _103 _51484 _9169873     _18053   _5   _1_24456   _24456   _178768  _30 _0 _0 _0 _0 _0 _40945
     58 SND     _132666  _225432  _904     _4820    _225432  _65536   _103 _51484 _9169873     _18053   _5   _1_24456   _24456   _178768  _36 _0 _0 _0 _0 _0 _40945
     59 ACK     _136771  _904     _194112  _0       _230252  _65536   _103 _51484 _9169873     _18053   _5   _1_24456   _24456   _178768  _40 _0 _0 _0 _0 _0 _40945
     60 ACK     _144761  _904     _196808  _0       _230252  _65536   _103 _36140 _9169873     _18053   _5   _0_24456   _24456   _194112  _28 _0 _0 _0 _0 _0 _40233
     61 ACK     _153778  _904     _196808  _0       _230252  _62976   _103 _33444 _9169873     _18053   _5   _0_24456   _24456   _196808  _26 _1 _0 _0 _1 _0 _39325
     62 SND     _153789  _196808  _904     _1348    _230252  _63488   _10  _33444 _1945318     _18053   _6   _1_24456   _24456   _196808  _26 _2 _15_0 _3 _2 _37160
     63 ACK     _158763  _904     _196808  _0       _230252  _63488   _10  _33444 _1945318     _18053   _6   _1_24456   _24456   _196808  _26 _2 _15_1 _3 _2 _37160
     64 SND     _158774  _198156  _904     _1348    _230252  _63488   _2   _33444 _2431647     _18053   _6   _0_24456   _24456   _196808  _26 _3 _23_1 _3 _2 _35778
     65 ACK     _171767  _904     _198156  _0       _230252  _63488   _2   _33444 _2431647     _18053   _6   _0_24456   _24456   _196808  _26 _3 _23_2 _3 _2 _35778
     66 ACK     _181819  _904     _199504  _0       _230252  _62464   _4   _32096 _2431647     _18053   _7   _1_24456   _24456   _198156  _25 _3 _22_1 _3 _2 _75805
     67 RTSND   _889192  _-100789098_-1348593670_0       _230252  _61440   _100 _30748 _1458988     _18053   _7   _0_0       _24456   _199504  _24 _3 _0 _0 _1 _2 _111455
     68 SND     _889202  _199504  _904     _1348    _230252  _61440   _1   _30748 _1458988     _18053   _7   _1_0       _24456   _199504  _24 _3 _21_0 _4 _2 _111455
     69 ACK     _913780  _904     _200852  _0       _230252  _61440   _1   _30748 _1458988     _18053   _7   _1_0       _24456   _199504  _24 _3 _21_1 _4 _2 _111455
     70 RTSND   _1721191 _-100789098_-1348593670_0       _230252  _60928   _100 _29400 _2431647     _18053   _8   _1_0       _24456   _200852  _23 _3 _0 _0 _1 _2 _142898
     71 SND     _1721201 _200852  _904     _1348    _230252  _60928   _1   _29400 _2431647     _18053   _8   _1_0       _24456   _200852  _23 _3 _20_0 _4 _2 _142898
     72 ACK     _1741783 _904     _202200  _0       _230252  _60928   _1   _29400 _2431647     _18053   _8   _1_0       _24456   _200852  _23 _3 _20_1 _4 _2 _142898
     73 RTSND   _2682200 _-100789098_-1348593670_0       _230252  _60928   _100 _28052 _2431647     _18053   _9   _1_0       _24456   _202200  _22 _3 _0 _0 _1 _2 _169911
     74 SND     _2682209 _202200  _904     _1348    _230252  _60928   _1   _28052 _2431647     _18053   _9   _1_0       _24456   _202200  _22 _3 _19_0 _4 _2 _169911
     75 ACK     _2701763 _904     _203548  _0       _230252  _60928   _1   _28052 _2431647     _18053   _9   _1_0       _24456   _202200  _22 _3 _19_1 _4 _2 _169911
     76 RTSND   _3642192 _-100789098_-1348593670_0       _230252  _60928   _88  _26704 _2431647     _18053   _10  _1_0       _24456   _203548  _21 _3 _0 _0 _1 _2 _193422
     77 SND     _3642204 _203548  _904     _1348    _230252  _60928   _1   _26704 _2431647     _18053   _10  _1_0       _24456   _203548  _21 _3 _18_0 _4 _2 _193422
     78 ACK     _3666780 _904     _204896  _0       _230252  _60928   _1   _26704 _2431647     _18053   _10  _1_0       _24456   _203548  _21 _3 _18_1 _4 _2 _193422
     79 RTSND   _4665208 _-100789098_-1348593670_0       _230252  _60928   _88  _25356 _2431647     _18053   _11  _1_0       _24456   _204896  _20 _3 _0 _0 _1 _2 _214619
     80 SND     _4665217 _204896  _904     _1348    _230252  _60928   _1   _25356 _2431647     _18053   _11  _1_0       _24456   _204896  _20 _3 _17_0 _4 _2 _214619
     81 ACK     _4684821 _904     _206244  _0       _230252  _60928   _1   _25356 _2431647     _18053   _11  _1_0       _24456   _204896  _20 _3 _17_1 _4 _2 _214619
     82 RTSND   _5690190 _-100789098_-1348593670_0       _230252  _60928   _88  _24008 _2431647     _18053   _12  _1_0       _24456   _206244  _19 _3 _0 _0 _1 _2 _232542
     83 SND     _5690199 _206244  _904     _1348    _230252  _60928   _1   _24008 _2431647     _18053   _12  _1_0       _24456   _206244  _19 _3 _16_0 _4 _2 _232542
     84 ACK     _5709796 _904     _207592  _0       _230252  _60928   _1   _24008 _2431647     _18053   _12  _1_0       _24456   _206244  _19 _3 _16_1 _4 _2 _232542
     85 RTSND   _6713191 _-100789098_-1348593670_0       _230252  _60928   _88  _22660 _2431647     _18053   _13  _1_0       _24456   _207592  _18 _3 _0 _0 _1 _2 _248224
     86 SND     _6713202 _207592  _904     _1348    _230252  _60928   _1   _22660 _2431647     _18053   _13  _1_0       _24456   _207592  _18 _3 _15_0 _4 _2 _248224
     87 ACK     _6737764 _904     _208940  _0       _230252  _60928   _1   _22660 _2431647     _18053   _13  _1_0       _24456   _207592  _18 _3 _15_1 _4 _2 _248224
 
     This may have nothing to do with bbr, but maybe you know this very well.
     Thanks.

Neal Cardwell

unread,
May 7, 2021, 9:12:15 AM5/7/21
to mingkun bian, BBR Development
Hi,

Thanks for posting.

A few questions to get a little more context/info:

+ What is the exact kernel version you are using? ("uname -r -v")
+ What's the congestion control algorithm you are using? ("sysctl net.ipv4.tcp_congestion_control")
+ What is your qdisc setup? ("tc qdisc show")
+ When you say, "we have changing the max rto to 1 second", can you please post any custom patches or sysctl settings you have made, either as attachments or a branch on github?
+ Can you please provide a tcpdump packet trace of this traffic? ("tcpdump -w /tmp/test.pcap -s 120 -i $DEVICE -c 100000 port $PORT")

thanks,
neal


--
You received this message because you are subscribed to the Google Groups "BBR Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bbr-dev/7b1b87cd-c927-4589-ac91-b60a55bae7adn%40googlegroups.com.

mingkun bian

unread,
May 8, 2021, 2:28:03 AM5/8/21
to BBR Development
Hi Neal,

+ What is the exact kernel version you are using? ("uname -r -v")
>>The original kernel version is “linux-4.18.0-147.el8”, I download source code from “https://access.redhat.com/downloads/content/479/ver=/rhel---8/8.1/x86_64/packages”,  then rebuild the kernel.

+ What's the congestion control algorithm you are using? ("sysctl net.ipv4.tcp_congestion_control")
>>net.ipv4.tcp_congestion_control = bbr

+ What is your qdisc setup? ("tc qdisc show")
>>
...
qdisc mq 0: dev enp5s0f1 root 
qdisc fq 0: dev enp5s0f1 parent :20 limit 10000p flow_limit 100p buckets 1024 orphan_mask 1023 quantum 3028 initial_quantum 15140 refill_delay 40.0ms 
qdisc fq 0: dev enp5s0f1 parent :1f limit 10000p flow_limit 100p buckets 1024 orphan_mask 1023 quantum 3028 initial_quantum 15140 refill_delay 40.0ms 
qdisc fq 0: dev enp5s0f1 parent :1e limit 10000p flow_limit 100p buckets 1024 orphan_mask 1023 quantum 3028 initial_quantum 15140 refill_delay 40.0ms 
qdisc fq 0: dev enp5s0f1 parent :1d limit 10000p flow_limit 100p buckets 1024 orphan_mask 1023 quantum 3028 initial_quantum 15140 refill_delay 40.0ms 
qdisc fq 0: dev enp5s0f1 parent :1c limit 10000p flow_limit 100p buckets 1024 orphan_mask 1023 quantum 3028 initial_quantum 15140 refill_delay 40.0ms 
...

+ When you say, "we have changing the max rto to 1 second", can you please post any custom patches or sysctl settings you have made, either as attachments or a branch on github?
>>
1. In tcp_timer.c->tcp_retransmit_timer, modification is as follows:
--- a/tcp_timer.c_bak
+++ b/tcp_timer.c
@@ -559,6 +559,9 @@ out_reset_timer:
        } else {
                /* Use normal (exponential) backoff */
                icsk->icsk_rto = min(icsk->icsk_rto << 1, TCP_RTO_MAX);
+               if (icsk->icsk_rto > 1*HZ) {
+                   icsk->icsk_rto = 1*HZ;
+               }
        }
        inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
                                  tcp_clamp_rto_to_user_timeout(sk), TCP_RTO_MAX);

2. another modification, I change the max zero window probes to 1 second.
In tcp.h->tcp_probe0_when
--- a/tcp.h_bak
+++ b/tcp.h
@@ -1233,6 +1233,9 @@ static inline unsigned long tcp_probe0_when(const struct sock *sk,
                                            unsigned long max_when)
 {
        u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
+        if(when > 1*HZ){
+            when = 1*HZ;
+        }
        return (unsigned long)min_t(u64, when, max_when);
 }

+ Can you please provide a tcpdump packet trace of this traffic? ("tcpdump -w /tmp/test.pcap -s 120 -i $DEVICE -c 100000 port $PORT")
>>
The current network traffic is relatively large, it may take a time to catch packet trace by tcpdump, I will provide it as quickly as possible.
Now the data sampled by bcc is basically the same as tcpdump,and there is more detailed information than tcpdump,I have uploaded the aligned data to the attachment, it may look better.  

For example:
"GET" means that an http GET request was received from client
"ACK" means that an tcp ACK was received  from client
"SND" means that server send body(size is not 0) or ACK(size is 0) to client 
"RTSND" means rto timeout, which enter "tcp_retransmit_timer", and next "SND" is the real send.

line 1, “GET” means that an http GET request was received from client, time is 0, seq and ack is 0, size is 822, rwnd(client_cwnd) is 83968 and so on.
line 2, "SND" means that server send  response to client, time is 10 us, seq is 0, ack is 822, size is 0(Server ack client's "GET")
line3,  "SND" means that server send  response to client, time is 3170 us, seq is 0, ack is 822, size is 5313
line4,  "ACK" means that an tcp ACK was received  from client.
...
line 67, "RTSND" means rto timeout, which enter "tcp_retransmit_timer", the seq and ack is not a real value.
line 68, "SND" means the real send which is marked by line 67.


bcc collect tcp stack and bbr info at the entrance of tcp_ack、tcp_v4_send_check and tcp_retransmit_timer,so some fields  need to update at next line,such as packets_out、sacked_out、lost_out、retrans_out.
tcptrace.txt

Neal Cardwell

unread,
May 8, 2021, 12:26:35 PM5/8/21
to mingkun bian, BBR Development, Yuchung Cheng
Thanks for the info!

I have a theory: for the RTSND event in line 67 (enterring tcp_retransmit_timer()), it logs:
   cwnd:100 ... lost_out:0 retrans_out:0 tcp_state:1 (TCP_CA_Disorder)

This suggests that at the time the RTO timer fired the flow was in TCP_CA_Disorder, with 0 packets marked as lost, and a big cwnd of 100. AFAICT this suggests that the previous ACK caused an undo event (probably based on TCP timestamps, aka an Eifel undo), which took the flow out of tcp_state:TCP_CA_Recovery and restored a large cwnd. Given that the flow was not in lost recovery, and did not think any packets were lost, that would explain why it's not retransmitting anything.

Hmm. My guess here is that the connection had reordering and then loss in the same flight of data, and triggered loss recovery, but when the connection detected the reordering  and undid the loss recovery, it was left without any RACK reordering timer in place for a quick detection of the real loss.

Upon an undo event, it seems if something is SACKed in the scoreboard we should schedule the RACK reorder timer based on the previously SACKed data, rather than falling back to an RTO as it seems we do here.

We will try to look into this to see if the loss recovery behavior can be improved.

Thanks for the report!

best,
neal

ps: In general, to know why packets were not sent, we'd need to know the cwnd and pacing rate and packets_in_flight before the connection started to decide whether to transmit something. So I would suggest that instead of instrumenting the entrance of tcp_ack(), it would shed more like to instrument the entrance of tcp_xmit_recovery(), which would show the state variables after the connection has updated its SACK scoreboard, cwnd, and pacing rate, slightly before the point at which it decides how much to send.

 

mingkun bian

unread,
May 10, 2021, 9:05:16 AM5/10/21
to BBR Development
Hi Neal,

    I have updated the more detailed bcc info(add tsval、tsecr and tcp_sock' recv_sack_cache[4] ) to attachment, and add more bcc samples.

    In tcp_ack,  tsecr = ((struct tcp_sock*)sk)->rx_opt.rcv_tsecr + ((struct tcp_sock*)sk)->tsoffset, but it does not look like  accurate sometimes.

    For example: 
    1. In tcptrace.txt: 
    In line 66, you say that "AFAICT this suggests that the previous ACK caused an undo event (probably based on TCP timestamps, aka an Eifel undo)",  and the tsecr is 2344362543(In SND lines, I can not find the same tsval, so I am not sure whether the calculation of tsecr is correct ), 
    In line 62, the  timestamp of the first retransmission is 2344362876, so 2344362543 < 2344362876, Eifel  will undo recovery.
 
    A. “Given that the flow was not in lost recovery, and did not think any packets were lost, that would explain why it's not retransmitting anything"
     >>In line 67, sndbuf is 30748, snd_una is 199504, snd_nxt is 230252, so there is not any more new date to send. assuming that there is still new data(sndbuf is bigger than 30748) to send at this time,  should new data be sent?

    B. “My guess here is that the connection had reordering and then loss in the same flight of data, and triggered loss recovery, but when the connection detected the reordering  and undid the loss recovery, it was left without any RACK reordering timer in place for a quick detection of the real loss.”
    >>According to the analysis of the above timestamp, Eifel undo the recovery in tcp_try_undo_partial,Is "without any RACK reordering timer" the logic of the red position?
    static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,
  bool is_dupack, int *ack_flag, int *rexmit)
    ...
     /* E. Process state. */
switch (icsk->icsk_ca_state) {
case TCP_CA_Recovery:
if (!(flag & FLAG_SND_UNA_ADVANCED)) {
if (tcp_is_reno(tp) && is_dupack)
tcp_add_reno_sack(sk);
} else {
if (tcp_try_undo_partial(sk, prior_snd_una))
return;
/* Partial ACK arrived. Force fast retransmit. */
do_lost = tcp_is_reno(tp) ||
  tcp_force_fast_retransmit(sk);
}
if (tcp_try_undo_dsack(sk)) {
tcp_try_keep_open(sk);
return;
}
tcp_identify_packet_loss(sk, ack_flag);
break;
     C. In line 67 - 87, after rto sending, why does  tcp state change to 1? In 67-69, Is it caused by the same reason Eifel(23443632732344363611)? 
          But it seems obvious that 69 line‘s "ACK" is the response of line 68' "SND".
 

     2. In  tcptrace2 .txt: 
      "TLPS" means that TLP send, which enter "tcp_send_loss_probe", the next line "SND" is the real send.
      In line 16,  the  timestamp of the first retransmission is 57722313
      In line 17, the ACK's tsecr is the same 57722313, It does not satisfy  Eifel algorithm,but why does it still undo recovery?Is it because of DSACK(line 17‘s recv_sack_cache will update in line 18) “0-1348”.
      In line 18 RTSND, tcp state is 1, but lost_out is 46 not 0,  it's different from tcprate.txt.

    3. In  tcptrace3 .txt:
     In line 1067, the starting tcp state of the problem is 4.

    4. In  tcptrace8 .txt:
     In line 403 404, the tsval and tsecr is the same 1250242242, Is this undo  caused by DSACK?


     Thanks
tcptrace6.txt
tcptrace3.txt
tcptrace8.txt
tcptrace.txt
tcptrace7.txt
tcptrace5.txt
tcptrace2.txt
tcptrace4.txt

Neal Cardwell

unread,
May 10, 2021, 8:59:40 PM5/10/21
to mingkun bian, BBR Development
On Mon, May 10, 2021 at 9:05 AM mingkun bian <bianm...@gmail.com> wrote:
Hi Neal,

    I have updated the more detailed bcc info(add tsval、tsecr and tcp_sock' recv_sack_cache[4] ) to attachment, and add more bcc samples.

I'm afraid that given this does not seem like a BBR bug, I don't have time to parse through all of the 8 traces you have provided. But I will try to offer some high-level thoughts.
 
    In tcp_ack,  tsecr = ((struct tcp_sock*)sk)->rx_opt.rcv_tsecr + ((struct tcp_sock*)sk)->tsoffset, but it does not look like  accurate sometimes.

    For example: 
    1. In tcptrace.txt: 
    In line 66, you say that "AFAICT this suggests that the previous ACK caused an undo event (probably based on TCP timestamps, aka an Eifel undo)",  and the tsecr is 2344362543(In SND lines, I can not find the same tsval, so I am not sure whether the calculation of tsecr is correct ), 
    In line 62, the  timestamp of the first retransmission is 2344362876, so 2344362543 < 2344362876, Eifel  will undo recovery.

So you can confirm that Eifel undo should happen in this case? Is this the same example with which you started this thread, or a new one? 
 
    A. “Given that the flow was not in lost recovery, and did not think any packets were lost, that would explain why it's not retransmitting anything"
     >>In line 67, sndbuf is 30748, snd_una is 199504, snd_nxt is 230252, so there is not any more new date to send. assuming that there is still new data(sndbuf is bigger than 30748) to send at this time,  should new data be sent?

If there is data in the sndbuf, and cwnd > packets_in_flight, then yes, generally something should be sent, unless pacing, the receive window, TSO deferral, TSQ, or Nagle algorithm prevent the transmission.
 
    B. “My guess here is that the connection had reordering and then loss in the same flight of data, and triggered loss recovery, but when the connection detected the reordering  and undid the loss recovery, it was left without any RACK reordering timer in place for a quick detection of the real loss.”
    >>According to the analysis of the above timestamp, Eifel undo the recovery in tcp_try_undo_partial,Is "without any RACK reordering timer" the logic of the red position?
    static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,
  bool is_dupack, int *ack_flag, int *rexmit)
    ...
     /* E. Process state. */
switch (icsk->icsk_ca_state) {
case TCP_CA_Recovery:
if (!(flag & FLAG_SND_UNA_ADVANCED)) {
if (tcp_is_reno(tp) && is_dupack)
tcp_add_reno_sack(sk);
} else {
if (tcp_try_undo_partial(sk, prior_snd_una))
return;
/* Partial ACK arrived. Force fast retransmit. */
do_lost = tcp_is_reno(tp) ||
  tcp_force_fast_retransmit(sk);
}
if (tcp_try_undo_dsack(sk)) {
tcp_try_keep_open(sk);
return;
}
tcp_identify_packet_loss(sk, ack_flag);
break;

There are two red positions, so I'm not sure which one you are referring to. But in addition to the issue you are pointing out, that losses would have been identified at a point in the code *after* the tcp_try_undo_partial() short-circuits, there is the issue that tcp_clean_rtx_queue() has the following code, which prevents RACK decisions based on ACKs matching spuriously-retransmitted sequence ranges:

                        if (!tcp_skb_spurious_retrans(tp, skb))
                                tcp_rack_advance(tp, sacked, scb->end_seq,
                                                 tcp_skb_timestamp_us(skb));

 
 
     C. In line 67 - 87, after rto sending, why does  tcp state change to 1?

The value 1 is TCP_CA_Disorder, so if the state is changing to TCP_CA_Disorder then this is because there are SACKed sequence ranges without the sender having decided that something is lost.
 
In 67-69, Is it caused by the same reason Eifel(23443632732344363611)? 
          But it seems obvious that 69 line‘s "ACK" is the response of line 68' "SND".

Sorry, I don't have time to parse that trace.
 
 

     2. In  tcptrace2 .txt: 
      "TLPS" means that TLP send, which enter "tcp_send_loss_probe", the next line "SND" is the real send.
      In line 16,  the  timestamp of the first retransmission is 57722313
      In line 17, the ACK's tsecr is the same 57722313, It does not satisfy  Eifel algorithm,but why does it still undo recovery?Is it because of DSACK(line 17‘s recv_sack_cache will update in line 18) “0-1348”.

If you are wondering why it undoes recovery, I would suggest instrumenting the undo-related functions (with "undo" in their name). DSACK would be one possible reason.
 
      In line 18 RTSND, tcp state is 1, but lost_out is 46 not 0,  it's different from tcprate.txt.

    3. In  tcptrace3 .txt:
     In line 1067, the starting tcp state of the problem is 4.

Sorry, I don't have time to parse that trace.
 
    4. In  tcptrace8 .txt:
     In line 403 404, the tsval and tsecr is the same 1250242242, Is this undo  caused by DSACK?

Here again, if you are wondering why it undoes recovery, I would suggest instrumenting the undo-related functions. DSACK would be one possible reason.

neal
 

Alexey Ivanov

unread,
May 10, 2021, 10:06:24 PM5/10/21
to Neal Cardwell, mingkun bian, BBR Development
Sorry for randomly jumping in. In case you are looking into possible
sources of network reordering I would suggest pinning your server
threads to CPUs (or trying enabling RFS in the kernel). In case your
are using Intel NICs you may consider disabling FlowDirector's ATR:
```
# ethtool --set-priv-flags eth0 flow-director-atr off
```

PS. We were recently troubleshooting a similar problem with
reordering, DSACKs, RACK heuristics, and Windows TCP stack[1].

[1] https://dropbox.tech/infrastructure/boosting-dropbox-upload-speed
> To view this discussion on the web visit https://groups.google.com/d/msgid/bbr-dev/CADVnQy%3DzK3szB3NKKPuNMyTVf91jSj%2Bqv-3KL4s%2BBc5dTp-VHA%40mail.gmail.com.



--
Sincerely,
Alexey Ivanov

mingkun bian

unread,
May 11, 2021, 9:39:00 AM5/11/21
to BBR Development
Hi Neal,
     I have updated three tcpdump packet traces and  corresponding bcc traces to  attachment,  which represent three different scenarios.
     1. tcptrace_10(undo because of DACK, then cyclic rto timeout): 
          a. In tcptrace_10.txt, in line 57, ack is 61405, and DSACK is 59977 - 61405(update in line 58), then undo recovery, but now all sndbuf data have sent,  
          and at this time, lost_out is 0, snd_una is 61405, snd_nxt is 66552, so now it can only wait rto timeout.
          At similar scene, in line 60,DSACK cause undo.
          b. Corresponding tcpdump is tcptrace_10.pacp, line is 58.

     2. tcptrace_13(undo because of  Eifel,  then cyclic rto timeout): 
        a. In tcptrace_13.txt, in line 1768, ack is 388225, packets_out is 17, sacked_out is 17, lost_out is 0(update in line 1769), this means that client have received all the data,
            but in line 1769 - 1779, server continue to send data(ack's tsecr is smaller than  timestamp of the first retransmission), so Eifel undo lost. it seems like that tcp core stack
            does not realize that all the data has been sent successly.
        b. Corresponding tcpdump is tcptrace_13.pacp, line is 1770.


     3. tcptrace_14(undo because of  DSACK,  then continue to send new data):
      a.  In tcptrace_14.txt, in line 115 - 116, DSACK cause undo,  but after line 127, line 130 can send new data.
      b. Corresponding tcpdump is tcptrace_14.pacp, line is 112.
tcptrace_14.pcap
tcptrace_10.txt
tcptrace_13.txt
tcptrace_14.txt
tcptrace_10.pcap
tcptrace_13.pcap

Yuchung Cheng

unread,
May 11, 2021, 12:45:04 PM5/11/21
to mingkun bian, BBR Development
Thank mingkun. I think the issue is as Neal has spotted that undo / Eifel could transition the state from Recovery -> undo -> Ca_disorder, and RACK was not engaged leading to timeout recovery. Neal and I will discuss internally how to address this as the code change is tricky with undo [undo is very complicated] and get back to you.

Message has been deleted
Message has been deleted
Message has been deleted
Message has been deleted

mingkun bian

unread,
May 13, 2021, 10:13:13 AM5/13/21
to BBR Development
Hi, 
    I hvae reproduced this problem through packetdrill(packetdrill_16.txt), and upload the tcpdump and bcc info.
    
    In tcptrace_16.txt:
    RTOSET means the entrance of tcp_rearm_rto.
    TLPSH means the entrance of tcp_schedule_loss_probe.
    RACKMK means the entrance of tcp_rack_mark_lost.
    RACKTM  means the entrance of tcp_rack_reo_timeout.(not sampled)

    1. In line 13, ACK is SACK(10000 - 11000), then RACK marked lost, lost_out is 10, packet_out is 11, sacket_out is 1.
    2. In line 15, retransmit one  packet(1-1000) because cwnd is 1(bbr_set_cwnd_to_recover_or_restore set cwnd 1).
    3. In line 16, ACK is 1000(DSACK is 0 - 1000),  in tcp_ack, tcp_set_xmit_timer will be executed,change the timer from RACK to RTO.

    static int tcp_ack(struct sock *sk, const struct sk_buff *skb, int flag)
    {
        ...
        /* If needed, reset TLP/RTO timer; RACK may later override this. */
        if (flag & FLAG_SET_XMIT_TIMER)
            tcp_set_xmit_timer(sk); 
    }

   Then exec tcp_fastretrans_alert's red position, undo_retrans is 0, undo_marker is not 0, so tcp_try_undo_dsa->cktcp_undo_cwnd_reduction will be executed, 
   then undo_marker is 0, cktcp_undo_cwnd_reduction return true, then tcp_try_keep_open will be executed, which change tcp_state from 3 to 1, but it does not 
   have change to exec tcp_identify_packet_loss, so now it can only wait for rto timeout(timer is set by RTO in tcp_ack's tcp_set_xmit_timer).

   static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,
  bool is_dupack, int *ack_flag, int *rexmit) {
    ...
    /* E. Process state. */
switch (icsk->icsk_ca_state) {
case TCP_CA_Recovery:
if (!(flag & FLAG_SND_UNA_ADVANCED)) {
if (tcp_is_reno(tp) && is_dupack)
tcp_add_reno_sack(sk);
} else {
if (tcp_try_undo_partial(sk, prior_snd_una))
return;
/* Partial ACK arrived. Force fast retransmit. */
do_lost = tcp_is_reno(tp) ||
  tcp_force_fast_retransmit(sk);
}
if (tcp_try_undo_dsack(sk)) {
tcp_try_keep_open(sk);
return;
}
tcp_identify_packet_loss(sk, ack_flag);
break;
    }

   4. In line 19-20, RTO timeout, then send packet(1000 - 2000)
   5. In line 21, ACK is DSACK(1000 - 2000), tcp_fastretrans_alert->tcp_process_loss will be executed, tcp_process_loss->tcp_try_undo_loss will be executed because snd_una have been updated.
      At the time, tp->undo_marker is not 0, undo_retrans is 0(because DSACK), so tcp_try_undo_loss->tcp_undo_cwnd_reduction will be executed, then lost_out is 0, undo_marker is 0, and tcp_state is 0(TCP_CA_Open) from 4.
      Then tcp_identify_packet_loss->tcp_rack_mark_lost->tcp_rack_detect_loss(line 24) will be executed, but now tp->tsorted_sent_queue is empty(because rack in line 13 have deleted the tcp_tsorted_anchors), so reo_timeout is 0,
      then it does not set rack timer, and can only wait rto timeout.

      At switch's default case, tcp_identify_packet_loss will be executed again(line 25), but tp->rack.advanced is 0, so it will do nothing.
      Then tcp_try_to_open will be executed(blue positon), it will change tcp_state from 0 to 1.
    
    static void tcp_fastretrans_alert(struct sock *sk, const u32 prior_snd_una,
  bool is_dupack, int *ack_flag, int *rexmit) {
    ...
/* E. Process state. */
switch (icsk->icsk_ca_state) {
case TCP_CA_Recovery:
                ...
break;
case TCP_CA_Loss:
tcp_process_loss(sk, flag, is_dupack, rexmit);
tcp_identify_packet_loss(sk, ack_flag);
if (!(icsk->icsk_ca_state == TCP_CA_Open ||
      (*ack_flag & FLAG_LOST_RETRANS)))
return;
/* Change state if cwnd is undone or retransmits are lost */
/* fall through */
default:
                ...
tcp_identify_packet_loss(sk, ack_flag);
if (!tcp_time_to_recover(sk, flag)) {
tcp_try_to_open(sk, flag);
return;
tcptrace_16.pcap
tcptrace_16.txt
packetdrill_16.txt

Yuchung Cheng

unread,
May 17, 2021, 8:32:15 PM5/17/21
to mingkun bian, BBR Development
Hi Mingkun,

Would you be able to try / test if we provide you a 4.18.0 kernel patch?

mingkun bian

unread,
May 17, 2021, 9:20:54 PM5/17/21
to BBR Development
Hi, 
    
    I would like to test it.
    
    Thanks.

Yuchung Cheng

unread,
May 18, 2021, 7:16:25 PM5/18/21
to mingkun bian, BBR Development
Great could you try this patch we have rebased to 4.18.0

0001-net-backports-net-TCP-better-handling-of-reordering-.patch

mingkun bian

unread,
May 18, 2021, 9:49:48 PM5/18/21
to BBR Development
  Hi, 
    I will test it, then report the result.
    Thanks.


在2021年5月18日星期二 UTC+8 上午8:32:15<Yuchung Cheng> 写道:

mingkun bian

unread,
May 19, 2021, 7:26:22 AM5/19/21
to BBR Development
Hi, 
    I have  incorporated the patch, but check successly by packetdrill(packetdrill_16.txt, tcp will occur rto)as mentioned  “https://groups.google.com/g/bbr-dev/c/OFHADvJbTEI/m/gZN5j9dbAwAJ” , because undo is caused by dsack(In tcp_try_undo_dsack) not  Eifel(In tcp_try_undo_partial),
         /* E. Process state. */
        switch (icsk->icsk_ca_state) {
        case TCP_CA_Recovery:
            if (!(flag & FLAG_SND_UNA_ADVANCED)) {
                    if (tcp_is_reno(tp) && is_dupack)
                            tcp_add_reno_sack(sk);
            } else {
                if (tcp_try_undo_partial(sk, prior_snd_una, &do_lost))
                    return;
            }
            if (tcp_try_undo_dsack(sk)) {
                    tcp_try_keep_open(sk);
                        return;
            }
            tcp_identify_packet_loss(sk, ack_flag);

            if (icsk->icsk_ca_state != TCP_CA_Recovery) {
                if (!tcp_time_to_recover(sk, flag))
                    return;
                /* Undo reverts the recovery state. If loss is evident,
                starts a new recovery (e.g. reordering then loss);             */
                tcp_enter_recovery(sk, (flag & FLAG_ECE));
           }
           break;

    So I delete a line in your patch as following(red position), which can check whether  loss after "dsack undo" is evident too, then packetdrill_16.txt check failed(tcp does not occur rto, but fastretrans)
         /* E. Process state. */
        switch (icsk->icsk_ca_state) {
        case TCP_CA_Recovery:
            if (!(flag & FLAG_SND_UNA_ADVANCED)) {
                    if (tcp_is_reno(tp) && is_dupack)
                            tcp_add_reno_sack(sk);
            } else {
                if (tcp_try_undo_partial_5(sk, prior_snd_una, &do_lost))
                    return;
            }
            if (tcp_try_undo_dsack(sk)) {
                    tcp_try_keep_open(sk);
                        //return;
            }
            tcp_identify_packet_loss(sk, ack_flag);

            if (icsk->icsk_ca_state != TCP_CA_Recovery) {
                if (!tcp_time_to_recover(sk, flag))
                    return;
                /* Undo reverts the recovery state. If loss is evident,
                starts a new recovery (e.g. reordering then loss);             */
                tcp_enter_recovery(sk, (flag & FLAG_ECE));
           }
           break;

  
    I use these two ways to build the kernel, which can switch different mode by /proc, then upgrade tomorrow night, and I will sample the info by tcpdump and bcc to compare the difference of those two ways.

    Thanks.

在2021年5月12日星期三 UTC+8 上午12:45:04<Yuchung Cheng> 写道:

Neal Cardwell

unread,
May 19, 2021, 9:24:14 AM5/19/21
to mingkun bian, BBR Development, Yuchung Cheng
Thanks. That approach for running the RACK logic after DSACK undo sounds reasonable to me. It is similar to one rev of an internal patch we considered during this process.

One related question: may we ask whether these traces are from synthetic workloads or real production user traffic? The DSACK behavior in the traces is quite strange, and seems to suggest either a bugger receiver TCP stack or a buggy middlebox.

For example, in tcptrace_10.pcap (time-sequence plot attached) the receiver sends DSACKs for many consecutive segments (one DSACKed segment in response to each RTO retransmit), without the sender having first seen cumulative ACKs or SACKs for those segments. That looks like buggy behavior from the receiver or some middlebox.

thanks,
neal



rack-dsack-tcptrace_10.pcap-time-seq-plot.png

mingkun bian

unread,
May 19, 2021, 11:14:40 PM5/19/21
to BBR Development
Hi Neal,
    Almost all the traces are real production user traffic except tcptrace_16(simulated by packetdrill,whose dest port is 8080).

    There are more scenarios like tcptrace10 because of DSACK without the sender having first seen cumulative ACKs or SACKs for those segments.
    As you said, that does not looks like sender's buggy , but it caused the slow speed of the tcp protocol stack.

Thanks.

Yuchung Cheng

unread,
May 20, 2021, 8:20:15 PM5/20/21
to mingkun bian, BBR Development
Mingkun,

Here is v2 patch to address this buggy receiver issue. I've verified it passes the (modified version) of your packetdrill.
However this patch is really a workaround to a receiver or middlebox problem, kinda hard to call it a "feature" and definitely not a fix for TCP loss recovery. 

0001-tcp-try-fast-recovery-after-spurious-recovery-due-to.patch
Message has been deleted

mingkun bian

unread,
May 22, 2021, 6:55:36 AM5/22/21
to BBR Development
Hi  Yuchung,

    I have upgraded the patch  to real production, issue(rto timeout because of undo from TCP_CA_Recovery to  TCP_CA_Disorder)   have  already solved, but  the rto timeout once per second caused by other reasons still exists.
    For example tcptrace_24.txt:
    line 42-45, lost_out is set 4 from 45, tcp_state is set 3 from 4:
    
    line 42, lost_out is 45,  undo_retrans is 11(updated in line 43), tcpstate is 4, retrans_out is 4(updated in line 43)
    line 43, RTSND means the entrance of tcp_retransmit_timer, then exec tcp_enter_loss->tcp_timeout_mark_lost->tcp_mark_skb_lost, it will set tp->retrans_out 0.
                  Then exec tcp_enter_loss->tcp_init_undo, then undo_retrans will be set -1(update in line 44).
    line 44, lost_out is 45(updated in line 45), undo_retrans is 1(updated in line 45), tcpstate is 4(updated in line 45), retrans_out is 1(updated in line 45, because resend a segment)
    line 45, lost_out is 4,(updated in line 46), undo_retrans is -1(updated in line 46), tcpstate is 3(updated in line 46), ackis 2697, sack is dsack(1349- 2697)(updated in line 46), retrans_out is 0(updated in line 46, becase dsack undo)

    /* E. Process state. */
switch (icsk->icsk_ca_state) {
case TCP_CA_Recovery:
                 ...
break;
case TCP_CA_Loss:
tcp_process_loss(sk, flag, is_dupack, rexmit);//ack is advanced, undo_retrans is 0(because of DSACK), so tcp_process_loss->tcp_try_undo_loss->tcp_undo_cwnd_reduction is executed, then lost_out is 0, tcp_state is TCP_CA_Open.
tcp_identify_packet_loss(sk, ack_flag);// it will marked new lost seg, then lost_out is 4
if (!(icsk->icsk_ca_state == TCP_CA_Open ||
      (*ack_flag & FLAG_LOST_RETRANS)))
return;
/* Change state if cwnd is undone or retransmits are lost */
/* fall through */
default:
                ...
if (icsk->icsk_ca_state <= TCP_CA_Disorder)
tcp_try_undo_dsack(sk);

tcp_identify_packet_loss(sk, ack_flag);
if (!tcp_time_to_recover(sk, flag)) {//tcp_time_to_recover will return true becase lost_out is 4.
tcp_try_to_open(sk, flag);
return;
}
                ...
/* Otherwise enter Recovery state */
tcp_enter_recovery(sk, (flag & FLAG_ECE));//tcp_state is set TCP_CA_Recovery from TCP_CA_Open


    line 61, tcp_state is TCP_CA_Disorder, snd_una is 9437, snd_nxt is 71445, rwnd is 62080, so it will not send a new segment becase of rwnd limit, it will only wait a rto timeout.
    line 62 - 92,  the rto timeout happen once per second, the reason is that tcp_state is set to 1 from 4
        /* E. Process state. */
switch (icsk->icsk_ca_state) {
case TCP_CA_Recovery:
                 ...
break;
case TCP_CA_Loss:
tcp_process_loss(sk, flag, is_dupack, rexmit);//ack is advanced, undo_retrans is 0(because of DSACK), so tcp_process_loss->tcp_try_undo_loss->tcp_undo_cwnd_reduction is executed,then lost_out is 0, tcp_state is TCP_CA_Open.
tcp_identify_packet_loss(sk, ack_flag);// no new segment is sent, so it does not mark new lost segment, lost_out is 0.
if (!(icsk->icsk_ca_state == TCP_CA_Open ||
      (*ack_flag & FLAG_LOST_RETRANS)))
return;
/* Change state if cwnd is undone or retransmits are lost */
/* fall through */
default:
                ...
if (icsk->icsk_ca_state <= TCP_CA_Disorder)
tcp_try_undo_dsack(sk);

tcp_identify_packet_loss(sk, ack_flag);
if (!tcp_time_to_recover(sk, flag)) {//tcp_time_to_recover will return false becase lost_out is 0
tcp_try_to_open(sk, flag);//tcp_state is set TCP_CA_Disorder from TCP_CA_Open, then it will not contine to retranmit data, only wait next rto timeout.
return;
}
    Thanks
tcptrace_21.txt
tcptrace_24.txt
tcptrace_23.pcap
tcptrace_22.pcap
tcptrace_21.pcap
tcptrace_22.txt
tcptrace_23.txt
tcptrace_24.pcap

Yuchung Cheng

unread,
May 24, 2021, 9:41:02 PM5/24/21
to mingkun bian, BBR Development


On Mon, May 24, 2021 at 6:37 PM Yuchung Cheng <ych...@google.com> wrote:
Hi Mingkun,

Thanks for testing the patch. We'll upstream that then.

For the new issue you've reported, unfortunately the problem is again the receiver / middlebox. The long recovery was due to this repeated behavior on a heavy loss situation:

1. TCP timeout and retransmit
2. Receiver acknowledges the retransmit with an DSACK indicating the retransmit was spurious
<rinse and repeat many times until every hole is RTO recoveried>

If these retransmits are all truly spurious, there should have been enough chances for a single cumulative ACK to ack all the original transmission. Not seeing such an ACK is a strong indication either the receiver or some middlebox is misbehaving. I do not think it's the sender's responsibility to further workaround this with more complexity. The undo logic is the most complicated part of TCP congestion control and has incurred dozens of bugs. Ultimately TCP needs a properly behavior receiver or middleboxes to perform well.

Hence I think it's the receiver or the network to address this issue, or take the penalty of slowing their users down.
image.png


--
You received this message because you are subscribed to the Google Groups "BBR Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+u...@googlegroups.com.

mingkun bian

unread,
May 25, 2021, 8:45:48 AM5/25/21
to BBR Development
Hi  Yuchung,
    We don't have permission to address the issue of receiver / middlebox, this is really not a problem with the tcp protocol stack, but it may affect our customer metrics.

    Two possible ways(it will increase the retransmission rate,  we will test them if the increase rate of retransmission is relatively low):
    1. Recognize this kind of scenario(send one segment per rto time)  by counting simply, then make tcp_try_undo_loss->tcp_undo_cwnd_reduction's second parameter false when certain conditions are met, then sender will retranmit more(tcpstate is 4 not 1) because lost_out is not 0.
    2. No longer reduce the count of retrans_out in tcp_mark_skb_lost, make sure undo_retrans not to set -1 in tcp_init_undo.

    Thanks.
Reply all
Reply to author
Forward
0 new messages