[ovs-dev] Support for LISP Control Plane

[Vina Ermagan (vermagan)]

On 6/4/13 3:07 PM, "Jesse Gross" <jesse at nicira.com> wrote:

>On Fri, May 31, 2013 at 11:50 AM, Vina Ermagan (vermagan)
><vermagan at cisco.com> wrote:
>> Hi All,
>>
>> This is a follow up on the discussion in late January about next steps
>>for
>> LISP support in OVS. We want to start executing on the medium term
>>plans,
>> namely support for LISP control plane (I copied the past conversation
>>below
>> for reference). We are interested in your feedback on this.
>>
>> We see two non-exclusive approaches to enable LISP cp in OVS:
>>
>> 1 ­ Use Open Flow to query the OF controller and install new mappings.
>>This
>> requires support for the flow-based tunneling
>> (action=set_field:<OVSx_IP>->tun_dst) in open flow protocol/controller.
>>
>> 2 ­ Extend vswitchd to support a light weight LISP cp, enabling
>>vswitchd to
>> query a (open source/commercial) LISP Map Server using the LISP protocol
>> (RFC 6830).
>>
>> We are thinking of starting with the second option, extending vswitchd
>>with
>> a light weight LISP cp. Thoughts?
>>
>> If this sounds good we can put together a more detailed plan on
>>extending
>> vswitchd, and share it sometime next week.
>
>Sorry, I was traveling last week and am just catching up on email.

No worries, thanks for your response.

>
>I think either approach is fine and, as you say, they aren't mutually
>exclusive. My only real concern with the second one is that we make
>sure that it is designed in such a way that it is modular - otherwise
>as we add support for more protocols the code will become difficult to
>maintain.

Absolutely right, modularity is a must. Great, then we will start looking
into it and will discuss the implementation plan in more detail on the
list before moving forward.

>
>Out of curiosity - I know that LISP has several more components in the
>dataplane that aren't implemented currently. Is the protocol useful
>without these or do you have plans to implement them?

Once the basic control plane is implemented, the protocol is useful for
sure. There are a few more flags on the dataplane defined in the RFC, but
most of those are rarely used. There are other components, such as
re-encapsulating tunnel routers (RTR) which can, for instance, be used for
NAT traversal; but the protocol is functional in many use cases  without
those components.  

Thanks,
Vina