On 28 Oct 2008, at 12:31, Thomas Schierl wrote:
Section 7.1.1.3 and 7.1.1.4 imply that a system must use the system real-time clock (Unix gettimeofday() or its equivalent on Windows) to generate the NTP-format clock used in RTCP SR packets. It then notes that this often has lower accuracy than desired. I don't see that requirement in the RTP specification: you need to use the same clock to generate the NTP wall-clock timestamps in all sessions to be synchronised, but RTP doesn't care what clock is used - you can use a different clock if that better meets your accuracy goals. Equally, I don't see any problem delaying an RTCP packet until the next clock tick, to get an accurate timestamp, if that's easier to implement. It might be useful to document implementation techniques somewhere, if there is evidence that SR timestamp are being widely mis-implemented. Cheers, -- Colin Perkins |
_______________________________________________ Audio/Video Transport Working Group avt at ietf.org https://www.ietf.org/mailman/listinfo/avt