[ovs-dev] [PATCH v2] ovn: Update TODO, ovn-northd flow table design, ovn-architecture for L3.

Han Zhou zhouhan at gmail.com
Thu Oct 8 17:51:02 UTC 2015


Hi Venkata,

On Thu, Oct 8, 2015 at 1:32 AM, Venkata Anil <vkommadi at redhat.com> wrote:
>
>
> On 10/08/2015 01:06 PM, Han Zhou wrote:
>>
>> Sorry, correct a typo:
>>
>> On Wed, Oct 7, 2015 at 11:55 PM, Han Zhou <zhouhan at gmail.com> wrote:
>>
>>> This would be a very interesting feature. But would it become scaling
>>> bottleneck if each HV need to be programmed to hold all the logical
>>> router flows (because it doesn't know whom could it be talking to)?
>
> I think already openstack neutron is following above approach.

Do you mean neutron DVR or any l3 service plugin? DVR uses linux
namespaces rather than openflow calculated by a controller. That is
completely different approach and yet no proof of large scale
deployment. If you mean some other service plugins please point out.
For example dragonflow is trying to implement it the reactive way I
mentioned: first crossing subnets packets are sent to controller.

Best regards,
Han



More information about the dev mailing list