Re: [pcp] Fwd: New Version Notification for draft-tsou-pcp-natcoord-10.txt

[Simon Perreault <[e-mail]>]

Le 2013-02-27 07:45, [e-mail] a écrit :
>> 1. The "P" bit comes from pcp-base's reserved range.  That bit is not
>> assignable to an option (layering violation).
>
> Med: Ok to move the parity bit to the option data field.

Just to be clear, this would increase the option length to 3 bytes, right?

Should we increase it to 4 for padding's sake?

>> 3. natcoord-10 says "If the Port Set Size is zero or one, a
>> MALFORMED_OPTION error is returned."  Disallowing a port set size of
>> one seems overly restrictive, especially when later in the same
>> section there is allowance that if the server cannot fulfill the
>> requested port set size, the server maps one port.  I agree a port
>> set size of 0 is nonsensical.
>
> Med: It does not make sense to include a PORT_SET to ask for one single port. The text you are referring to is when the port set size > 1.

Well, it is clear that PORT_SET size of 1 is of little usefulness, but 
it can make sense if you just focus on the semantics. The are the same 
as a PORT_SET-less MAP request.

Usually having two ways of saying the same thing is bad(TM). I'm curious 
as to why Dan thinks it should still be allowed.

>> 5. "If a mapping already exists and the PORT_SET option can be honored"
>> could also benefit from some editing to aid clarity.  For
>> example, let's
>> say a single port is mapped (internal port = 5000), and then PORT_
>> SET is used with internal port=6000.  It needs to be clearer
>> what happens
>> there -- is the new request 'honored' by destroying the previous
>> mapping and creating this new one, or is only a single port mapped or
>> something else?  Related, need a discussion in the document of what
>> happens if PORT_SET is used and a subsequent PCP request adjusts
>> the lifetime of a single port within that port set.
>
> Med: I guess you want to say PORT_SET with internal port=5000, no? An example may be added to the text to aid clarity.

Also assuming internal port=5000, I think we need more than an example. 
There's undefined behaviour here.

>> 6. In Section 4.3, Port Set Renewal and Deletion, it is unclear how
>> PORT_SET changes the behavior for the Assigned External IP Address
>> versus what is already in
>> http://tools.ietf.org/html/draft-ietf-pcp-base-29#page-32.  If the
>> change is intentional, it should be highlighted clearer in natcoord;
>> if it is re-stating what is already in pcp-base, that should be
>> clarified, too (or simply removed).
>
> Med: Can you explicit what changes are you referring to? Thanks.

I don't understand either. An example would be nice.

>> 7. I think the PORT_SET option should be in the non-mandatory
>> to process range.  Because the fall back (for when the server
>> doesn't understand PORT_SET) is exactly in line with what existing
>> code would have to handle anyway if the PORT_SET cannot be
>> fulfilled.
>
> Med: Works for me.

Me too.

Simon