Re: [straw] Review of draft-ietf-straw-b2bua-rtcp-10

[Lorenzo Miniero <[e-mail]>]

On Fri, 27 May 2016 17:17:35 -0500
"Ben Campbell" <[e-mail]> wrote:

> [Oops, I sent that from the wrong address. Please send any discussion
> to [e-mail]]
> 
> Hi,
> 
> Here is my review of draft-ietf-straw-b2bua-rtcp-10. I mentioned in
> my AD evaluation of a previous version that the draft needed editing
> for clarity and readability. In my opinion, this version is improved,
> but still a bit rough in places. I have a few substantive comments
> and quite a few editorial comments.
> 
> This is not a formal AD evaluation. That will occur after the chairs 
> submit a new publication request. But I think we need to work through 
> this before it will be ready for IETF last call.
> 


Hello again, Ben,

I just posted a new version of the document (an announcement should get
to the ML soon) that tries to address most, if not all, of your points.
You can find some additional comments inline on the parts I haven't
addressed yet (or I only partly addressed), for one reason or another.
To make this more readable, I snipped the parts that I already
integrated.


> 
> * The intended status is proposed standard. Do people really expect 
> b2bua vendors to implement the rules in this draft? Do they expect 
> existing b2bua implementations to be updated?
> 


I already answered to this in my previous mail, but just to reinstate,
I'll leave the decision to WG and chairs/ADs as to what the intended
status should be. I personally would keep it a proposed standard, but
again wouldn't object to a BCP, if needed.


> 
> * Section 3, last paragraph - This paragraph seems to argue that the 
> fact that
>    existing b2buas do not follow specs is a reason to create more
> specs. That seems
>    questionable.
> 


Fair point. I haven't changed that part in the doc yet, as I wanted to
be sure I'd get it right, first. Do you think it would be ok to say
something along the lines of "due to the mixed nature of B2BUAs, they
sometimes can't or won't implement all specifications, so they should
at least do this"?


> * Section 3.1:
> 
>      * Second Paragraph
>           * "but having them synced during setup would
>             help nonetheless" - Help how? Please elaborate, or
> consider dropping the text in the parentheses.
> 


This came up in a STRAW meeting some time ago. The main scenario for
that is WebRTC usage, where browsers typically advertize the SSRCs
they're going to use in advance. The RTCWEB RTP usage drafts,
specifically, mandates support for signalled SSRC identifiers (see
Section 4.8), although it also clarifies that endpoints should be
prepared to handle SSRCs that were not advertized. The text in the
parenthesis wanted to refer to that kind of usage, but I realize the
wording was pretty awful, so I dropped it as you suggested.


> 
> * Section 4
>     * Last paragraph, 2nd to last sentence - If you think people will 
> really pay attention,
>       then this might be a good place for a MUST or SHOULD
> 


Do you mean where we say "behaviour is unspecified and discouraged"? Or
more in general about the fact that a B2BUA can only follow the
guidelines when it has access to editing fields without breaking
hashes? Not sure which MUST or SHOULD we might add there, considering
that, if the media the B2BUA is encrypted, it can't possibly edit RTP
related information without disrupting the communication, which in
turn means no RTCP editing will be needed/possible either.

Do you think a "B2BUAs MUST NOT modify RTP/RTCP data if media is
encrypted" to be enough, in case?


> 
> * Section 1, First Paragraph:
> 
>      * I'm not sure what "... can lead to unexpected situations the
>        IETF is trying to address" means.


It was an awkward attempt at referring to the scope of the STRAW WG
itself. I removed the text after "situations".


> 
>      * Third paragraph:
>          * "It SHOULD NOT, though, blindly forward all SDP
> attributes," 
> - Isn’t
>            the idea that it should forward any attributes that it 
> doesn’t
>            have a specific reason not to forward? (i.e. default to 
> forwarding?)
> 


The issue here is the reason not to forward itself. In order to figure
out whether an attribute can be forwarded or not, the B2BUA must have
some knowledge of it and of its impact. What I wanted to highlight in
that sentence is that there are attributes that, if left where they are
by a B2BUA unaware of their purpose, can actually cause issues. Just
think, for instance, of 'rtcp' attribute that is explained in the text,
or other attributes supporting features a B2BUA may not support itself.

Of course, as you point out there could be reasons to actually keep
attributes you don't know about in there, e.g., because they advertize
support for an end-to-end feature the B2BUA doesn't need to worry
about. Besides, the attributes that can cause issues are probably not
that many anyway.

I'll think about a way to make this clearer in the next for next
version.


Thanks!
Lorenzo