[ovs-discuss] HW virt functions support

Jesse Gross jesse at nicira.com
Wed Feb 1 00:07:38 UTC 2012


On Tue, Jan 31, 2012 at 3:42 PM, potworyk <potworyk at gmail.com> wrote:
> 2012/2/1 Jesse Gross <jesse at nicira.com>:
>> On Tue, Jan 31, 2012 at 3:13 PM, potworyk <potworyk at gmail.com> wrote:
>>> 2012/1/31 Jesse Gross <jesse at nicira.com>:
>>>> On Tue, Jan 31, 2012 at 2:27 PM, potworyk <potworyk at gmail.com> wrote:
>>>>> Hi,
>>>>>
>>>>> I've searched for information about hardware support in Open vSwitch
>>>>> but there is no clear statment. Does Open vSwitch use SR-IOV, Intel
>>>>> VMDq or other? Where I can find some information about it?
>>>>
>>>> Those types of hardware offloads don't have the type of fine-grained
>>>> control that OVS needs to operate.
>>>
>>> Wouldn't that improve performance? For example pre sorting packets
>>> (and interrupts)  based on L2 data to specific CPU's, then on each
>>> core perform the upper layer checking?
>>
>> Open vSwitch supports flow-based control at a level that is much more
>> fine-grained than L2.  There's no guarantee that the L2 information
>> alone will tell you where the packet needs to go.
>>
>> RSS spreads processing and interrupts based on a hash, which is supported.
>
> But in a specific situation, for example when there is a flow entry in
> openflow table which defines only vlan or mac address (in my opinion
> common situation) this could be optimization. This would allow to
> achieve line rate performance and still allow to use openflow for
> advanced features.

Sure, it can be used to optimize certain situations.  Please go ahead
and submit your patches and associated performance numbers to the dev
mailing list when you are ready.


More information about the discuss mailing list