Re: [AVT] I-D ACTION:draft-ietf-avt-rtcp-non-compound-02.txt

[Randell Jesup <rjesup at wgate.com>]

"Ingemar Johansson S" <ingemar.s.johansson at ericsson.com> writes:
>The observant reader might notice that the more advanced bandwidth
>computation algorithms given in the previous draft has been removed in
>the updated draft. A motivation to this is given in the draft text. I
>have tried to find out what kinds of issues that non-compound RTCP might
>cause for SRTP so if anybody has info regarding this I would be very
>grateful.

SRTP/SRTCP should be unaffected by this change.  

>>Section 6.1: a simple test that non-compound packets are not rejected  
>>out-of-hand might be to send a non-compound SR packet, then check  
>>that the timestamp is echoed back in the corresponding RR packet?
>A first glance this sounds like a neat solution. It may not cover the
>exotic case that a middle box accepts non-compound RTCP with only SR
>packet but rejects all other types of non-compound RTCP.

It does sound useful.

>>Section 6.2.1: the first four bullet points are unnecessary, since  
>>non-compound packets can be detected by comparing their size with the  
>>size of the containing transport layer packet.
>I guess this boils down to what the exact definition of what a
>non-compound RTCP is. If the non-compound RTCP contains only one RTCP,
>then you are right. However the current definition in the draft is that
>a non-compound RTCP is one that deviates from the minimum (or full)
>compound RTCP rules in RFC3550. Perhaps the only wise thing is to narrow
>down the definiton such that a non-compound RTCP is one that contains
>one RTCP, no matter the type, this would simply the discrimination.

That sounds reasonable to me (since the restriction we're trying to relax
is the "must have SR/RR plus one other" rule).  The other bullet points
sound mostly like heuristics to *guess* whether it's non-compound without
actually looking at what's in it - an odd thing to do, IMHO.  The sole
exception would be if you need to know if it's a non-compound SRTCP packet
*without* decrypting it (since only the length of the first packet is in
the clear.  Why you'd want to know that, I don't know.

Note: the "size of first RTCP sub-packet plus overhead == UDP packet size"
test is NOT correct unless you also account for SRTCP overhead.  If you
want to keep that description, please add some phrase about including
SRTCP overhead (which may vary depending on what was negotiated for tag
length and optional MKI).

Note that section 9.1 of RFC 3550 will be affected by this change, though
I don't think that will be any problem.  It may be worth mentioning.

-- 
Randell Jesup
rjesup at wgate.com
"The fetters imposed on liberty at home have ever been forged out of the weapons
provided for defence against real, pretended, or imaginary dangers from abroad."
		- James Madison, 4th US president (1751-1836)

_______________________________________________
Audio/Video Transport Working Group
avt at ietf.org
http://www.ietf.org/mailman/listinfo/avt