Severe IPv6 TCP transfer issues on 13.0-RC1 and RC2
Marek Zarychta
zarychtam at plan-b.pwste.edu.pl
Tue Mar 16 14:18:48 UTC 2021
W dniu 16.03.2021 o 12:50, tuexen at freebsd.org pisze:
>> On 16. Mar 2021, at 11:55, Blake Hartshorn <contact at blakehartshorn.com> wrote:
>>
>> Hi Michael,
>>
>> I've attached tcpdumps for port 80 on both sides of a bad transfer, using 2 VMs in the same datacenter, FreeBSD 13 serving and 12 as a client. A friend of mine suggested I also run some tests with iperf3, so pasting those results below. You'll see it going fast in one direction and crawling in the other on TCP. There's also some disparity on UDP.
> The problem is that the server provides TCP segments larger than the MTU
> to the NIC. These are dropped and needs to be retransmitted. That is why
> it takes so long. So I guess TSO is enabled on the NIC and not working correctly.
>
> What is the output of ifconfig? Can you disable TSO? Does that work around
> the problem?
>
> Best regards
> Michael
>>
>> TCP:
>> [ ID][Role] Interval Transfer Bitrate Retr Cwnd
>> [ 5][TX-C] 0.00-1.00 sec 27.7 MBytes 233 Mbits/sec 342 26.7 KBytes
>> [ 7][RX-C] 0.00-1.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 1.00-2.00 sec 15.8 MBytes 132 Mbits/sec 249 52.0 KBytes
>> [ 7][RX-C] 1.00-2.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 2.00-3.00 sec 13.7 MBytes 115 Mbits/sec 307 15.4 KBytes
>> [ 7][RX-C] 2.00-3.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 3.00-4.00 sec 14.5 MBytes 121 Mbits/sec 260 22.4 KBytes
>> [ 7][RX-C] 3.00-4.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 4.00-5.00 sec 14.3 MBytes 120 Mbits/sec 240 37.9 KBytes
>> [ 7][RX-C] 4.00-5.00 sec 5.58 KBytes 45.7 Kbits/sec
>> [ 5][TX-C] 5.00-6.00 sec 17.7 MBytes 149 Mbits/sec 363 15.4 KBytes
>> [ 7][RX-C] 5.00-6.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 6.00-7.00 sec 14.8 MBytes 124 Mbits/sec 287 8.38 KBytes
>> [ 7][RX-C] 6.00-7.00 sec 5.58 KBytes 45.7 Kbits/sec
>> [ 5][TX-C] 7.00-8.00 sec 14.7 MBytes 123 Mbits/sec 293 28.1 KBytes
>> [ 7][RX-C] 7.00-8.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 8.00-9.00 sec 11.9 MBytes 100 Mbits/sec 325 18.3 KBytes
>> [ 7][RX-C] 8.00-9.00 sec 4.18 KBytes 34.3 Kbits/sec
>> [ 5][TX-C] 9.00-10.00 sec 14.3 MBytes 120 Mbits/sec 315 39.3 KBytes
>> [ 7][RX-C] 9.00-10.00 sec 4.18 KBytes 34.3 Kbits/sec
>> - - - - - - - - - - - - - - - - - - - - - - - - -
>> [ ID][Role] Interval Transfer Bitrate Retr
>> [ 5][TX-C] 0.00-10.00 sec 159 MBytes 134 Mbits/sec 2981 sender
>> [ 5][TX-C] 0.00-10.00 sec 159 MBytes 134 Mbits/sec receiver
>> [ 7][RX-C] 0.00-10.00 sec 77.0 KBytes 63.1 Kbits/sec 65 sender
>> [ 7][RX-C] 0.00-10.00 sec 44.6 KBytes 36.6 Kbits/sec receiver
>> -----------------------------------------------------------------------------------------------------------
>>
>>
>> UDP:
>> [ ID][Role] Interval Transfer Bitrate Jitter Lost/Total Datagrams
>> [ 5][TX-C] 0.00-1.00 sec 81.6 MBytes 685 Mbits/sec 67798
>> [ 7][RX-C] 0.00-1.00 sec 8.80 MBytes 73.8 Mbits/sec 0.255 ms 54070/60475 (89%)
>> [ 5][TX-C] 1.00-2.00 sec 72.7 MBytes 610 Mbits/sec 64802
>> [ 7][RX-C] 1.00-2.00 sec 8.52 MBytes 71.5 Mbits/sec 0.154 ms 68912/75116 (92%)
>> [ 5][TX-C] 2.00-3.00 sec 73.7 MBytes 618 Mbits/sec 64158
>> [ 7][RX-C] 2.00-3.00 sec 8.52 MBytes 71.5 Mbits/sec 0.276 ms 67738/73945 (92%)
>> [ 5][TX-C] 3.00-4.00 sec 76.6 MBytes 643 Mbits/sec 63521
>> [ 7][RX-C] 3.00-4.00 sec 8.55 MBytes 71.8 Mbits/sec 0.160 ms 68647/74874 (92%)
>> [ 5][TX-C] 4.00-5.00 sec 76.1 MBytes 638 Mbits/sec 64614
>> [ 7][RX-C] 4.00-5.00 sec 8.55 MBytes 71.7 Mbits/sec 0.461 ms 67542/73767 (92%)
>> [ 5][TX-C] 5.00-6.00 sec 75.9 MBytes 637 Mbits/sec 64834
>> [ 7][RX-C] 5.00-6.00 sec 8.57 MBytes 71.9 Mbits/sec 0.297 ms 71565/77806 (92%)
>> [ 5][TX-C] 6.00-7.00 sec 73.0 MBytes 613 Mbits/sec 63639
>> [ 7][RX-C] 6.00-7.00 sec 8.40 MBytes 70.5 Mbits/sec 0.199 ms 69545/75663 (92%)
>> [ 5][TX-C] 7.00-8.00 sec 74.6 MBytes 626 Mbits/sec 65030
>> [ 7][RX-C] 7.00-8.00 sec 8.78 MBytes 73.6 Mbits/sec 0.254 ms 67173/73566 (91%)
>> [ 5][TX-C] 8.00-9.00 sec 75.0 MBytes 629 Mbits/sec 64848
>> [ 7][RX-C] 8.00-9.00 sec 8.77 MBytes 73.5 Mbits/sec 0.298 ms 70932/77315 (92%)
>> [ 5][TX-C] 9.00-10.00 sec 74.5 MBytes 625 Mbits/sec 64487
>> [ 7][RX-C] 9.00-10.00 sec 8.71 MBytes 73.1 Mbits/sec 0.185 ms 68268/74612 (91%)
>> - - - - - - - - - - - - - - - - - - - - - - - - -
>> [ ID][Role] Interval Transfer Bitrate Jitter Lost/Total Datagrams
>> [ 5][TX-C] 0.00-10.00 sec 754 MBytes 632 Mbits/sec 0.000 ms 0/647731 (0%) sender
>> [ 5][TX-C] 0.00-10.12 sec 105 MBytes 87.2 Mbits/sec 0.245 ms 571090/647649 (88%) receiver
>> [ 7][RX-C] 0.00-10.00 sec 1009 MBytes 846 Mbits/sec 0.000 ms 0/761013 (0%) sender
>> [ 7][RX-C] 0.00-10.12 sec 86.2 MBytes 71.4 Mbits/sec 0.185 ms 674392/737139 (91%) receiver
>>
>>
Taking a look at this iperf output I recalled that 80% of my setups
suffered from similar issue after transitioning from 1{1,2}-STABLE to
13-STABLE about a mounth ago. I have asked on IRC but nobody confirmed
similar problems so I have reduced MTU to 8900 from the original 9000 on
some vlan(4) interfaces to solve the issue. I am using mostly vlan(4)s
over LACP lagg(4)s created on NICs. So far (for FreeBSD 11 and 12)
setting MTU 9000 on physical NIC was sufficient to make it work, now I
have set MTU 9000 on NICs and reduced MTU 8900 on vlan(4)s.
>>
>>
>> On Tuesday, March 16, 2021 4:16:15 AM EDT tuexen at freebsd.org wrote:
>>>> On 15. Mar 2021, at 12:56, Blake Hartshorn <contact at blakehartshorn.com> wrote:
>>>>
>>>> The short version, when I use FreeBSD 13, delivering data can take 5 minutes for 1MB over SSH or HTTP when using IPv6. This problem does not happen with IPv4. I installed FreeBSD 12 and Linux on that same device, neither had the problem.
>>>>
>>>> Did some troubleshooting with Linode, have ultimately ruled the network itself out at this point. When the server is on FreeBSD 13, it can download quickly over IPv6, but not deliver. Started investigating after noticing my SSH session was lagging when cat'ing large files or running builds. This problem even occurs between VMs in the same datacenter. I generated a 1MB file of base64 garbage served by nginx for testing. IPv6 is being configured by SLAAC and on both 12 and 13 installs was setup by the installer. Linode uses Linux/KVM hosts for their virtual machines so it's running on that virtual adapter.
>>>>
>>>> I asked on the forums, another user recommended going to the mailing lists instead. Does anyone know if config settings need to be different on 13? Did I maybe just find a real issue? I can provide any requested details. Thanks!
>>> Could you prove a .pcap tracefile, one from the sender, one from the recevier, of
>>> a TCP/IPv6 connection, which doesn't work as expected. For example, use your 1MB
>>> base64 garbage transfer.
>>>
>>> Best regardes
>>> Michael
>>>>
>>>>
>>>> _______________________________________________
>>>> freebsd-net at freebsd.org mailing list
>>>> https://lists.freebsd.org/mailman/listinfo/freebsd-net
>>>> To unsubscribe, send any mail to "freebsd-net-unsubscribe at freebsd.org"
>>>
>>>
>> <freebsd12-client.pcap.gz><freebsd13-server.pcap.gz>
>
> _______________________________________________
> freebsd-net at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-net
> To unsubscribe, send any mail to "freebsd-net-unsubscribe at freebsd.org"
>
--
Marek Zarychta
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/freebsd-net/attachments/20210316/1cf30985/attachment.sig>
More information about the freebsd-net
mailing list