On 11/16/2011 09:36 AM, Colin Perkins wrote:
[catching up]
On 31 Oct 2011, at 08:50, Harald Alvestrand wrote:
On 10/30/2011 02:38 PM, Justin Uberti wrote:
In Appendix A, minor error in the diagram for the extension. The second 0xBE should be 0xDE. 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | 0xBE | 0xBE | length=1 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | ID | len=2 | send timestamp (t_i) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ For the REMB message, what do we expect the "SSRC of packet sender" header field to be set to? Is it just an arbitrary choice of any of the SSRCs used by the sender of the REMB? Apparently one of Magnus' new drafts goes into that territory. I suspect Colin Perkins will have opinions too.
I think that for efficiency and general sanity, these need to be sent from one of the SSRCs and not from all of the SSRCs. I don't see any reason not to pick one at random.
I can imagine there are scenarios where you'd want to congestion control each stream individually, and have a receiver partition up its available bandwidth estimate between them. In that case, you'd send per-SSRC.
If you want to control a group of streams, then signalling a single SSRC, and making assumptions about which other SSRCs it applies to, doesn't make sense to me. It would be clearer, and more future proof, to define an identifier for the group of streams, signal that in SDP, and report that identifier in the feedback messages.
That is, be explicit, rather than implicitly trying to group things based on assumptions about which set of SSRCs should be controlled together. As described, the feedback option explicitly lists the SSRCs it applies to:
0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |V=2|P| FMT=15 | PT=206 | length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | SSRC of packet sender | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | SSRC of media source | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Unique identifier 'R' 'E' 'M' 'B' | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Num SSRC | BR Exp | BR Mantissa | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | SSRC feedback | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | ... | The discussion has only focused on the "SSSRC of packet sender" field. The timestamp option diagrammed at the beginning of this thread will, of course, only apply to the packet it occurs in.