Re: [lisp] LISP Overlay Model

[Dino Farinacci <[e-mail]>]

<no hats on>
> 
>> On 25 Aug 2015, at 18:27, Dino Farinacci <[e-mail]> wrote:
>> 
>> I agree with everything Fabio stated. To me this means, in totality, we support the following data-planes and corresponding well-known port numbers:
>> 
>> (1) L3 LISP ala RFC 6830, port 4341
>> (2) L2 LISP ala Smith Internet Draft, port 8472
>> (3) VXLAN already in the field, port 4789
>> (4) LISP-GPE, port 4341
> 
> I do not think this would work. 
> How can you make the difference between a LISP-GPE and a LISP header when both use the same UDP port?

There really isn’t any difference. The P-bit is cleared in the RFC 6830 (because it is unspecified). I was being explicit above to not exclude LISP-GPE.

> You would need side information about whether or not an xTR supports or not the LISP-GPE header.

There is compatability text in the LISP-GPE draft to handle this. But that is why I proposed the Encapsulation Format Type LCAF, so an ITR knows the data-plane formats the ETR supports. We’ll need to depend on this to support multiple data-planes.

> Feasible, but I fear that we will end up with so many corner cases that will make the solution complex.

Well I agree 100%. And I would push for just (1) and (2) only. But people will object. 

So if you really want to be practical (which I do want to be), we should support (1) and (3) only. Because that is what is already deployed in the field. (1) for L3 overlays, and (2) for L2 overlays.

The data-plane situation in NVo3 is a total mess with no adult supervision. I have to state this because we need to be careful how much mess we bring into the LISP WG.

> Further, LISP-GPE and VXLAN-GPE are so similar that makes me wonder why should we have both?

This discussion has already occurred. You need the authors to justify this.

Dino