[FFmpeg-devel] [PATCH] Sync the NTP timestamps for RTP streams

[Luca Barbato]

On 03/08/2010 11:52 AM, Luca Abeni wrote:
> Hi Martin,
> 
> Martin Storsj? wrote:
>> Hi,
>>
>> As discussed earlier, the NTP timestamps in RTCP sender reports are
>> used to sync multiple RTP streams together. If first_rtcp_ntp_time in
>> RTP muxers differ, though, the individual streams get out of sync.
> [...]
> 
> [...]
>> The attached patch series is an attempt at this. First the ntp_time
>> function is made public, since the caller (be it the RTSP muxer or an
>> external libavformat user) will need it to set coordinated NTP start
>> timestamps. Is this ok, or should it be ff_ prefixed and left internal
>> to libavformat for now?
> 
> I am not sure if we want a public av_ntp_time() function. I'd be
> inclined to go for a private ff_ntp_time(), but I'd like to hear
> other people's opinions too.

I think it could have some uses outside, yet it's quite simple so either
way won't cause much harm.

>> Then the RTP muxer is made to check for a metadata with the key
>> "ntp_start_time", and set the first_rtcp_ntp_time with the value in
>> that key, if present. Is this acceptable, or should another field be
>> added to AVFormatContext for this?
> 
> I like very much this kind of solution (and I think that adding
> muxer-dependent fields to AVFormatContext is not a good idea). Maybe
> we can use a prefix for the metadata to indicate that it is a parameter
> for the RTP muxer (something like "RTP/ntp_start_time", or similar).
> I expect we will need other metadata entries like this one (for allowing
> setting the SSRC, the RTP timestamp offset, etc...).
> If this is not considered an abuse of the metadata API (I like it, but
> I'd like to hear Michael's opinion on this), patch 0002 is ok.

I think it's nice as well, if Michael finds that ugly we could just make
a parameters api for this purpose leveraging the experience with the
metadata api...

>> Lastly, the RTSP muxer uses this to coordinate the start NTP time for
>> all RTP sessions. Is adding a field to RTSPState for this ok? This
>> needs to be set once for the whole RTSP session, but is needed deep
>> within the call stack, in rtsp_rtp_mux_open, so passing it through as
>> an argument isn't awkward. Another option would be to store it in
>> metadata for the RTSP AVFormatContext.
> 
> No particular opinions on this. Looks ok to me, but it's better to see
> what Ronald and Luca B. think about it.

I think is ok as well. Ronald?

lu

-- 

Luca Barbato
Gentoo/linux
http://dev.gentoo.org/~lu_zero