Re: [rtcweb] Plan A, respun - bundle-only attribute

[Cullen Jennings <[e-mail]>]

Hmm - most of this seems to just be the same as bundle. At least that was how I read the draft. 


On May 8, 2013, at 12:37 PM, Dale R. Worley <[e-mail]> wrote:

> This is a promising proposal, but I think a lot more care needs to be
> taken in defining how port numbers are to be used in this technique.
> I can't tell what particular uses of port numbers are allowed and what
> ones are not allowed.  And it seems that the intended patterns could
> result in problematic behaviors from intermediate devices.
> 
> Erratum:  You state "you can always have more than one bundle", but
> the example in section 6 does not contain an "a=group" attribute.
> 
> It seems that an offered bundle consists of:
> 
> - one or more m= lines offering non-zero ports
> 
> - zero or more m= lines offering zero ports but with "a=bundle-only"
> 
> Question:  Is it true that there must be at least one m= line with a
> non-zero port,

There must be at least one is how I read the draft when coupled with current bundle draft 

> or can *all* the m= lines be bundle-only (if the UA is
> willing to have none of the media streams if the other UA doesn't
> support bundling)?  If all may have zero ports, what port is to be
> used for transport?
> 
> Question:  If there is more than one m= line offering non-zero ports,
> are identical port numbers required/permitted/forbidden?  It would
> appear from backward-compatibility considerations that the answer is
> "forbidden", but that is not stated.
> 
> An answered bundle consists of:
> 
> - zero or more m= lines that are rejected by containing a zero port
> 
> - one or more m= lines that are accepted by containing a non-zero port
> 
> Question:  If there is more than one m= line offering non-zero ports,
> are identical port numbers required/permitted/forbidden?  You state
> that the answer is "permitted", but that seems problematic:

Uh, I read this as it is the bundle draft 

> Question:  It is clear what the intended transport association is if
> both the offer and answer contain only one non-zero port number.  What
> is the intended transport association(s) in other cases?

whatever bundle says

> 
> Question:  It is possible that the only accepted m= lines in the
> bundle are ones that were offered with zero ports and
> "a=bundle-only".  This results in an intended packet flow that does
> not correspond to the offered/answered port for any single one of the
> m= lines.  This may cause problems with intermediate devices.

Seems like the clean up O/A in bundle would clean this up - so same as bundle 

> 
> Question:  If several m= lines are offered with different ports and
> they are accepted (with the same port), what is the intended transport
> association?  It seems that in any case, there will be an m= line with
> an offered/answered port pair that will not receive any traffic, which
> may cause problems with intermediate devices.

This is the same as bundle

> 
> Question:  There can be an m= line offered with a zero port that is
> answered with a non-zero port.  This contravenes the legacy
> offern/answer rules and may cause problems with intermediate devices.
> 
> As an additional matter, there is this statement:
> 
>   An old endpoint simply rejects the bundle-only m-lines by responding
>   with a 0 port.  (This isn't a normative statement, just a description
>   of the way the older endpoints are expected to act.)
> 
> Why is this characterized as "isn't a normative statement"?  By legacy
> offer/answer rules, the endpoint is *required* to respond to an
> offered zero port with a zero port.
> 
> Dale
> _______________________________________________
> rtcweb mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rtcweb