Date: Wed, 11 Jan 2017 12:13:20 -0800 From: Matthew Macy <mmacy@nextbsd.org> To: "Sean Bruno" <sbruno@freebsd.org> Cc: =?UTF-8?Q?=22Olivier_Cochard-Labb=C3=A9=22?= <olivier@freebsd.org>, "freebsd-net@freebsd.org" <freebsd-net@freebsd.org>, "freebsd-current@freebsd.org" <freebsd-current@freebsd.org> Subject: Re: HEADS-UP: IFLIB implementations of sys/dev/e1000 em, lem, igb pending Message-ID: <1598f29d379.ea6360351471.8752933472741761813@nextbsd.org> In-Reply-To: <6c6cb534-73c7-464b-8af1-7445a9c0188c@freebsd.org> References: <30f21c75-d3a2-edcd-1999-d5ed9f970c06@freebsd.org> <b000a957-8d17-a04d-6275-0d3920aa8a17@freebsd.org> <CA%2Bq%2BTcramTrYgYT-s%2B=aBZzRJV8FmKQqGt=1twPhLBR7AoXkcQ@mail.gmail.com> <1598d97bf2a.c6bcb76838987.6501340920645175463@nextbsd.org> <574a7ac7-4842-9518-8286-a4d89a9f7a27@freebsd.org> <CA%2Bq%2BTco-dcoU8EZnDEzgoK-v2Q2=U5GF6ASMSj0kwzd_wB5xig@mail.gmail.com> <6c6cb534-73c7-464b-8af1-7445a9c0188c@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
---- On Wed, 11 Jan 2017 12:02:06 -0800 Sean Bruno <sbruno@freebsd.org> wr=
ote ----=20
> =20
> =20
> On 01/11/17 12:47, Olivier Cochard-Labb=C3=A9 wrote:=20
> > On Wed, Jan 11, 2017 at 4:17 PM, Sean Bruno <sbruno@freebsd.org=20
> > <mailto:sbruno@freebsd.org>> wrote:=20
> > =20
> > =20
> > =20
> > Olivier:=20
> > =20
> > Give this a quick try. This isn't the correct way to do this, but=
I=20
> > want to see if I'm on the right path:=20
> > =20
> > =20
> > =E2=80=8Bthanks, it fix the problem, I've got back the 4 queues:=E2=80=
=8B=20
> > =20
> > =E2=80=8Bigb2: <Intel(R) PRO/1000 PCI-Express Network Driver> port 0x3=
000-0x301f=20
> > mem 0xdfea0000-0xdfebffff,0xdff24000-0xdff27fff irq 18 at device 20.0 =
on=20
> > pci0=20
> > igb2: attach_pre capping queues at 8=20
> > igb2: using 1024 tx descriptors and 1024 rx descriptors=20
> > igb2: msix_init qsets capped at 8=20
> > igb2: pxm cpus: 4 queue msgs: 9 admincnt: 1=20
> > igb2: using 4 rx queues 4 tx queues=20
> > igb2: Using MSIX interrupts with 5 vectors=20
> > igb2: allocated for 4 tx_queues=20
> > igb2: allocated for 4 rx_queues=20
> > igb2: Ethernet address: 00:08:a2:09:33:da=20
> > igb2: netmap queues/slots: TX 4/1024, RX 4/1024=20
> > =E2=80=8B=20
> > In forwarding mode, I measure about 10% performance drop with this new=
=20
> > drivers on this hardware:=20
> > =20
> > x head r311848: packets per second=20
> > + head r311849 and BAR patch: packets per second=20
> > +---------------------------------------------------------------------=
-----+=20
> > |++ ++ + xxx =
x x|=20
> > | |__M__A__=
__| |=20
> > | |___AM__| =
|=20
> > +---------------------------------------------------------------------=
-----+=20
> > N Min Max Median Avg S=
tddev=20
> > x 5 924170 943071 927509 931612.1 8096=
.8269=20
> > + 5 831452 845929.5 840940 838730.5 6413=
.5602=20
> > Difference at 95.0% confidence=20
> > -92881.6 +/- 10652.2=20
> > -9.96999% +/- 1.07481%=20
> > (Student's t, pooled s =3D 7303.85)=20
> > =20
> > Regards,=20
> > =20
> > Olivier=20
> > =20
> =20
> =20
> Hmmm ... did your old tests do 4 or 8 queues on this hardware?=20
> =20
> Did the old tests run 1024 tx/rx slots or the max 4096?=20
That's a great point, only having one thread per core could easily account =
for this. I'm hoping Sean can make txq !=3D rxq work so that you can have 8=
txqs and 4 rxqs.
-M
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1598f29d379.ea6360351471.8752933472741761813>