Question about DPDK performance on bare-metal VS virtual

[Christian Ehrhardt]

Hi,
we benchmark that stuff as well, but your setup is quite close to "what
everybody else uses" which for the good side of it should (tm) work :-)
It more or less matches the most commonly used test I've seen like
https://networkbuilders.intel.com/docs/open-vswitch-enables-sdn-and-nfv-transformation-paper.pdf
I have tried some different setups with less focus on "just" packet
forwarding and some more classic workloads instead and was able to identify
issues as well.
Over time once I'm sure this will become an upstream discussion for
guidance.

I hope I can help your case with a few tweaks below.

On Mon, Apr 11, 2016 at 8:56 AM, Martinx - ジェームズ <thiagocmartinsc at gmail.com>
wrote:

>
>
> On 11 April 2016 at 03:36, Martinx - ジェームズ <thiagocmartinsc at gmail.com>
> wrote:
>
>> Guys,
>>
>>  Lets get a server, with 2 x 1G NIC embedded, and 2 x 10G NIC igbxe
>> (compatible with DPDK + PMD).
>>
>> [...]

> ovs-vsctl add-br br1 -- set bridge br1 datapath_type=netdev
> ovs-vsctl add-port br1 dpdk1 -- set Interface dpdk1 type=dpdk
> ovs-vsctl add-port br1 vhost-user2 -- set Interface vhost-user2
> type=dpdkvhostuser
>

By default that will use just one PMD per port which IMHO is nice and
correct if you expect a lot of ports with a small share each.
But if you have just these and drive A LOT over them you should consider:

1. enabling multiple rx queues for these ports - this is done via e.g.:
   ovs-vsctl set Open_vSwitch . other_config:n-dpdk-rxqs=4
2. enable multiqueue when connecting the guest to it (match the queues you
set up before)

All of that is already in the development branch of the new serverguide.

 BTW, is it possible to translate those qemu options, into a Libvirt XML?
> Very annoying to deal with those commands...
>

Yes, again I listed that already in the new serverguide.

I think for now you should really just do
bzr branch lp:serverguide serverguide-16.04-dev
cd serverguide-16.04-dev
make serverguide-html


Even with that I cannot guarantee any hard performance numbers, but I hope
that helps.
Have fun and I'm sure after this you will hit the next issue, so let us
know.


Christian Ehrhardt
Software Engineer, Ubuntu Server
Canonical Ltd
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/ubuntu-server/attachments/20160411/591e3900/attachment.html>