Date: Tue, 12 Apr 2022 16:08:29 -0400 From: Matt Garber <matt.garber@gmail.com> To: "Patrick M. Hausen" <hausen@punkt.de> Cc: Charles Sprickman <spork@bway.net>, FreeBSD-STABLE Mailing List <freebsd-stable@freebsd.org>, Kristof Provost <kp@freebsd.org>, mike tancsa <mike@sentex.net> Subject: Re: vtnet rxcsum broken for forwarding RELENG_13 ? Message-ID: <CANwXMPMp7ziaVFD2G6VfpbAooTJGizuwaHNuf43vEMrR%2B4YELg@mail.gmail.com> In-Reply-To: <745890A5-983D-41BA-9592-D283EE800AD6@punkt.de> References: <d30a54ad-6b93-456e-64fc-75d1b09b2fb3@sentex.net> <CANwXMPPUEYWOoYLcYGhzMpP=MOd-oNrT4S7NJUy8AE52cPRvEg@mail.gmail.com> <0FE1F488-EEA5-4010-9926-2D9567E8461F@FreeBSD.org> <5A9B449D-BC3C-4D89-8AE8-7CC680B2F41E@bway.net> <322649DF-446E-4BAE-876D-D4FC47FE84B0@FreeBSD.org> <745890A5-983D-41BA-9592-D283EE800AD6@punkt.de>
next in thread | previous in thread | raw e-mail | index | archive | help
--0000000000007d340c05dc7aa05c Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable On Tue, Apr 12, 2022 at 4:01 PM Patrick M. Hausen <hausen@punkt.de> wrote: > Hi Kristof, hi all, > > > Am 12.04.2022 um 21:48 schrieb Kristof Provost <kp@FreeBSD.org>: > > That PF checksum issue was fixed > c110fc49da2995d10d60d908af0838ecb4be9bee, back in 2015. > > I still have abysmal performance with pf NAT in a DigitalOcean droplet > running 13.1-RC2 unless I configure: > > ifconfig_vtnet0=3D"-rxcsum -txcsum -rxcsum6 -txcsum6" > > I can give you SSH access, if needed. > > Kind regards, > Patrick Same for me, on 12.x RELEASEs, and I=E2=80=99d previously tested on Digital= Ocean and Google Compute Platform infrastructure. While I don=E2=80=99t doubt tha= t some issues with TCP checksums have potentially been resolved, there are still unresolved performance problems using the vtnet driver (VirtIO, KVM host) unless checksums are disabled. This might only be specific to NAT/forwarding at this point, as I realized my setup also involved PF NAT=E2=80=99ing on cloned loopback interfaces for non-VNET jailed services. Also note that other KVM setups using drivers other than vtnet don=E2=80=99= t seem to have the same problem, at least based on the alternatives I tested =E2= =80=94 e.g., AWS Nitro KVM instances are unaffected as they=E2=80=99re using the E= lastic Network Adapters. Thanks, =E2=80=94Matt --0000000000007d340c05dc7aa05c Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable <div>On Tue, Apr 12, 2022 at 4:01 PM Patrick M. Hausen <<a href=3D"mailt= o:hausen@punkt.de">hausen@punkt.de</a>> wrote:<br></div><div><div class= =3D"gmail_quote"><blockquote class=3D"gmail_quote" style=3D"margin:0px 0px = 0px 0.8ex;border-left-width:1px;border-left-style:solid;padding-left:1ex;bo= rder-left-color:rgb(204,204,204)">Hi Kristof, hi all,<br> <br> > Am 12.04.2022 um 21:48 schrieb Kristof Provost <kp@FreeBSD.org>:= <br> > That PF checksum issue was fixed c110fc49da2995d10d60d908af0838ecb4be9= bee, back in 2015.<br> <br> I still have abysmal performance with pf NAT in a DigitalOcean droplet<br> running 13.1-RC2 unless I configure:<br> <br> =C2=A0 =C2=A0 =C2=A0 =C2=A0 ifconfig_vtnet0=3D"-rxcsum -txcsum -rxcsum= 6 -txcsum6"<br> <br> I can give you SSH access, if needed.<br> <br> Kind regards,<br> Patrick</blockquote><div dir=3D"auto"><br></div><div dir=3D"auto">Same for = me, on 12.x RELEASEs, and I=E2=80=99d previously tested on Digital Ocean an= d Google Compute Platform infrastructure. While I don=E2=80=99t doubt that = some issues with TCP checksums have potentially been resolved, there are st= ill unresolved performance problems using the vtnet driver (VirtIO, KVM hos= t) unless checksums are disabled.</div><div dir=3D"auto"><br></div><div dir= =3D"auto">This might only be specific to NAT/forwarding at this point, as I= realized my setup also involved PF NAT=E2=80=99ing on cloned loopback inte= rfaces for non-VNET jailed services.</div><div dir=3D"auto"><br></div><div = dir=3D"auto">Also note that other KVM setups using drivers other than vtnet= don=E2=80=99t seem to have the same problem, at least based on the alterna= tives I tested =E2=80=94 e.g., AWS Nitro KVM instances are unaffected as th= ey=E2=80=99re using the Elastic Network Adapters.</div><div dir=3D"auto"><b= r></div><div dir=3D"auto"><br></div><div dir=3D"auto">Thanks,</div><div dir= =3D"auto">=E2=80=94Matt</div><div dir=3D"auto"><br></div><div dir=3D"auto">= <br></div><div dir=3D"auto"><br></div><blockquote class=3D"gmail_quote" sty= le=3D"margin:0px 0px 0px 0.8ex;border-left-width:1px;border-left-style:soli= d;padding-left:1ex;border-left-color:rgb(204,204,204)" dir=3D"auto"></block= quote></div></div> --0000000000007d340c05dc7aa05c--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CANwXMPMp7ziaVFD2G6VfpbAooTJGizuwaHNuf43vEMrR%2B4YELg>