Re: [lisp] Gen-ART IETF Last Call review of draft-ietf-lisp-ddt-08

[Dino Farinacci <[e-mail]>]

> I reviewed draft-ietf-lisp-ddt-08, and my memory is that the only
> significant technical question was regarding the "D" bit in Map-Request
> messages.

Let me try to make this more clear for you Dale. Thanks for the comment.

> Thinking back on it, I believe that the difficulty I was having was with
> the explanation of the D bit, not its functionality.  In particular, a
> DDT server can share a node and a listening port with a Map-Server.

Yes, every Map-Server that is part of a child referral from a DDT-node is also a DDT-node. But the DDT-node does not process D=0 Map-Requests but does D=1 Map-Requests by responding with Map-Referral messages. A Map-Server sends Map-Referrals too. And usually a D=0 Map-Request is forwarded by Map-Servers to ETRs so they send Map-Replies.

> Both of these servers process Map-Request messages, albeit with
> different semantics.  Hence the D bit in Map-Request messages is needed
> to differentiate which server is to process a given Map-Request message.

The reason I explained the above was that the D-bit tells the receiver of a Map-Request what type of message to return regardless of the colocation status of the servers.

> My suggestion is that the document define the D bit in this way.  Once
> that is done, it's obvious for a particular client sending a particular
> message whether the D bit should be set.  Compare with the current

Well depending on what type of device you are referring to as a “client”. Map-Resolvers are part of infrastructure and they are typically the only devices that send D=1 Map-Requests. And they are sending them because they know the Map-Request is going to a DDT node and that they want a Map-Referral as a response.

> description, which speaks of the D bit as if it is a classification of
> the sender of the Map-Request, which leads to conceptual problems

It is not the classification of the sender.

> because it requires the document to define the classification and ensure
> that all possible clients are properly classified.
> 
> (Ideally, I would advocate that a DDT Map-Request should have a
> different type code than a Map-Server Map-Request.  But I'm sure that it
> is too late to put that fix into existing deployments.)

They both have the exact same information, one just solicits a Map-Referral and D=0 doesn’t.

Dino