Re: [pcp] WG Call for Adoption: draft-tsou-pcp-natcoord-09

[Dave Thaler <[e-mail]>]

I have read the draft and all of the emails from others regarding it.

From the list discussion, there does appear to be rough consensus for
having the ability for an application to get a range of ports in one
PCP message exchange rather than requiring a separate exchange per port.

That said, I do not think that draft-tsou-pcp-natcoord-09 is ready to be
adopted as a WG draft.

First, it seems clear that there are a number of use cases for which a range
of ports is useful (as Simon, Sam, and Ian among others have
pointed out).  However, draft-tsou-pcp-natcoord is too heavily focused
on one particular scenario, both in terms of motivation text and in terms
of the packet format proposed).  And it seems it's still unclear
(as Alain and Dan have stated) whether PCP is the right solution for
that scenario.  For the more generic use cases, I personally believe the
specific packet format proposed is not a good match as noted below.

Specific feedback:

1) Personally I find the use of port set prefix/suffix/index/mask
   confusing and unnecessarily complex for the more general scenarios
   of "please give me N ports".  For a general mechanism I would instead
   expect a simple "number of ports desired/allocated".  That simple
   mechanism would also be more flexible.  For example, it could allow
   sizes that are not a power of two, thus preventing wasted ports in
   some scenarios.

2) I believe a new Opcode is not the right solution for requesting a
   range of ports.  Instead, I think it would be more appropriate to have
   an Option, usable with the existing MAP Opcode, that had the number
   of ports desired/allocated starting at the external port number already
   present in the MAP Opcode.  This allows the existing logic and constraints
   regarding MAP to be applied without duplicating it in another spec
   or in another set of code.

3) The motivation text does not talk about other use cases, and indeed
   I'd prefer that a document we adopt be generic in supporting port
   range scenarios and not focus on a use case on which there isn't clear
   consensus whether PCP is the right solution.

4) The title and filename are not particularly helpful in conveying what
   the document is about.  I'd expect a document whose title and filename
   are about a port range extension.

If the above items were done in a doc, it's possible the result might have
little or no text from the current document, and I wonder whether it'd
be faster to start from scratch.  Hence I don't think it's ready to be
adopted as a WG draft yet.

-Dave