[openstack-dev] [quantum] SR-IOV VNIC support in quantum

Isaku Yamahata yamahata at valinux.co.jp
Fri Feb 22 15:47:40 UTC 2013


On Thu, Feb 21, 2013 at 07:28:41PM +0100, Ian Wells wrote:
> It's certainly something we're looking at (in Cisco) at the moment.  I
> absolutely think it's worth having a session, though there are several parts to
> the problem, and it crosses the nova-quantum divide which can make it awkward
> to schedule:
> 
> - SR-IOV and directmap into the hypervisor (we're looking at Zardara Storage's
> work on this with libvirt, which has never been incorporated into the mainline
> but basically works fine)
> - scheduling based on resources that are limited in number - that is,
> scheduling to a machine that has remaining unallocated SRIOV NICs (or whatever
> else)
> - how one might get an SRIOV NIC into a Quantum network

Assuming that NIC has features like anti-spoofing/vlan tag insertion in
NIC l2 switching logic(I suppose intel NIC has such features),
Quantum networking can work with OpenFlow top-of-rack switch.

Probaly the following blue print can be enhanced to support SR-IOV
Generic Hardware Driver interface for Quantum Plugins
https://blueprints.launchpad.net/quantum/+spec/ovsplugin-hardware-devices

thanks,

> And of course, not all devices that you might wish to directmap are NICs.
> 
> Migration's not something I've been giving much thought to, I admit.
> --
> Ian.
> 
> 
> On 21 February 2013 15:45, Jiang, Yunhong <yunhong.jiang at intel.com> wrote:
> 
>     Hi, all
>             I want to know if there are any plan to support SR-IOV networking
>     in openstack?
> 
>             SR-IOV network reduce the CPU utilization greatly and improve the
>     scalability and sharing capability greatly [1]. Combining with bonding
>     driver [2], it can also support live migration.
> 
>             However, SR-IOV networking also have some issue traditionally. For
>     example, it's not easy to control flow from SR-IOV NIC, and not easy to
>     setup the tenant network. However, SDN can ease such limitation greatly,
>     also some private environment does not have highly control requirement and
>     will be helpful.
> 
>             Any idea? Can this be a topic candidate in design summit?
> 
>     Thanks
>     --jyh
> 
>     [1] http://static.usenix.org/event/wiov08/tech/full_papers/dong/dong.pdf
>     [2] http://www.kernel.org/doc/ols/2008/ols2008v2-pages-261-267.pdf
> 
>     _______________________________________________
>     OpenStack-dev mailing list
>     OpenStack-dev at lists.openstack.org
>     http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> 
> 

> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


-- 
yamahata



More information about the OpenStack-dev mailing list