Basically, you can map each virtual nic card to one VM and the physical interface is the slave of bridge interface of OVS. By default, the offload is accelerated, so some how you can switch between the accelerated path and OVS path.
On Fri, Mar 21, 2014 at 5:51 AM, Jayanthi Annadurai <jayant...@gmail.com>wrote: > Our goal is to offload the datapath to the SR-IOV NIC which has switching > capability. > And for some flows, the packet has to be routed to OVS datapath > (userspace/kernel). > Basically, I am trying to use both OVS and vswitch hardware acceleration. > Is this > possible with SR-IOV NIC? > > > > > > > > On Thu, Mar 20, 2014 at 12:04 PM, Wes Felter <w...@felter.org> wrote: > >> On 3/18/14 11:35 AM, Jayanthi Annadurai wrote: >> >>> Introduction: >>> I am working on adding hardware acceleration support in OVS. The >>> hardware has multiple ethernet ports, SRIO-V and switching capability. >>> >>> Issue: >>> How to add these physical ports to OVS bridge? I am trying to >>> understand, whether to connect all the physical ports as a single port >>> to OVS bridge or not. >>> >>> Can someone suggest how to handle this case? >>> >> >> There's some interesting work going on in SnabbSwitch using SR-IOV to get >> incoming packets onto the right core and then using a vSwitch to perform >> its normal processing. It seems like this optimization could also be used >> in OVS. >> >> The part that I don't understand is how to get SR-IOV to work with LAG. >> >> -- >> Wes Felter >> IBM Research - Austin >> >> _______________________________________________ >> discuss mailing list >> discuss@openvswitch.org >> http://openvswitch.org/mailman/listinfo/discuss >> > > > _______________________________________________ > discuss mailing list > discuss@openvswitch.org > http://openvswitch.org/mailman/listinfo/discuss > >
_______________________________________________ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
