Hi Avi,
There may be a terminology issue here:
-vhost-user interfaces are used between VPP and QEMU for fast communications.
It doesn't involve Linux kernel at all.
-in VPP 18.01, we've added support for virtionet (tap v2 interfaces). In this
case VPP "plays" the usual role of QEMU. This new interfaces are quite
efficient. In this model, skbuff created by Linux kernel cannot be mapped to
user space. However, Linux makes it possible to create skbuff where data is
located in the mmaped memory region. As a result, packets created by VPP do not
require copy to be injected in the kernel.
-Anyway, packet copy is not a real bottleneck here. We did an experimental
patch where VPP support GRO/GSO with tapv2. 2 containers were running iperf and
the bottleneck was tcp stack...
Jerome
Le 14/02/2018 09:08, « Avi Cohen (A) » <avi.co...@huawei.com> a écrit :
Thank you Jarome
I'm working with LXC , but should be applied for docker as well
I can connect the container through a virtio-user port in VPP, and a tap
interface in kernel
But we pay for a vhost kthread that is copying data from kernel to user
space and viceversa.
Another option is to connect with veth pair - but the performance is
further degraded
Another issue is how VPP interface with sandbox?
Best Regards
Avi
> -----Original Message-----
> From: vpp-dev@lists.fd.io [mailto:vpp-dev@lists.fd.io] On Behalf Of Jerome
> Tollet
> Sent: Tuesday, 13 February, 2018 11:27 PM
> To: vpp-dev@lists.fd.io
> Cc: vpp-dev@lists.fd.io
> Subject: Re: [vpp-dev] VPP vs XDP-eBPF performance numbers - for linux
> containers
>
> Hi Avi,
> Can you elaborate a bit on the kind of containers you'd like to run.
> Interfaces exposed to containers may be different if you are looking to
run
> regular endpoints vs cVNF.
> Jerome
>
> Le 13/02/2018 15:04, « vpp-dev@lists.fd.io au nom de Avi Cohen (A) » <vpp-
> d...@lists.fd.io au nom de avi.co...@huawei.com> a écrit :
>
> Hello
> Are there 'numbers' for performance - VPP vs XDP-eBPF for container
> networking.
>
> Since the DPDK and linux-containers are not compatible, is a sense
that
> container and host share the same kernel - hence pkts received at VPP-DPDK
> at user-space and directed to a linux container - should be go down to
the
> kernel and then to the container ip-stack, while in XDP-eBPF this pkt can
be
> forward to the container ip-stack directly from the kernel.
>
> I heard that a vhostuser interface for containers is 'in-working'
stage.
> Can anyone assist with the performance numbers and the status of this
> vhost-user for containers ?
>
> Best Regards
> Avi
>
>
>
>
>
>
>
-=-=-=-=-=-=-=-=-=-=-=-
Links:
You receive all messages sent to this group.
View/Reply Online (#8208): https://lists.fd.io/g/vpp-dev/message/8208
View All Messages In Topic (4): https://lists.fd.io/g/vpp-dev/topic/11144798
Mute This Topic: https://lists.fd.io/mt/11144798/21656
New Topic: https://lists.fd.io/g/vpp-dev/post
Change Your Subscription: https://lists.fd.io/g/vpp-dev/editsub/21656
Group Home: https://lists.fd.io/g/vpp-dev
Contact Group Owner: vpp-dev+ow...@lists.fd.io
Terms of Service: https://lists.fd.io/static/tos
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub
-=-=-=-=-=-=-=-=-=-=-=-
