hija,!
On 05/20/2011 02:35 PM, Peder Hedlund wrote:
> The obvious guess would be that your CPU is too slow and can't keep
> up with recoding the video stream.
> The other guess is that the camera and/or the interface can't push
> that amount of data in real time to the PC.
well, i'm using 3G quadcore, and the CPU usage is quite low. i haven't
tested the spikes when doing more action and zooms.
Robin Gareus <robin(a)gareus.org>rg>:
The problem is likely the former (the DV input is a
bitstream with
constant bps and usually no problem): Panning and zooming puts
addidtional some strain on the encoder (motion compensation calc) and
also produces bandwidth spikes.
AT LAC, we were able to kick out some remote-participants with quick
camera panning due to bandwith spikes so we tried to avoid those :)
BTW, we were able to max-out CPU a 2.8GHz quad-core with two
ffmpeg2theora instances.
You can check if decreasing the quality (`-v 3`) helps and give
`--speedlevel 2` a try: it uses less CPU, for increased output
bandwidth.
Down-scaling the video decreases CPU usage of the encoder, but
requires a few CPU cycles to perform the scaling, in particular if
scaling to non multiples of 16: try `-x 384 -y 288 -aspect 4:3`
instead of ` -x 393 -y 288` (it /should/ not make a difference since
Theora requires the geometry to be multiples of 8 (or 16?), but
ffmpeg2theora may just be scaling/cropping/expanding twice?!).
good points. thanks.
i was especially eager to hear ideas from the experienced LAC team!
it is true we have very low upload bandwidth here (256k). we will
upgrade it in next few days to 768k. when i lowered resolution to
160x128 it seems to help quite a lot...
i will do more tests, but the conclusion seems to be that the zooms and
many actions creates bandwidth spikes and that causes skips between
encoder and server, correct?
l.
--
sujet est machinique!
Nova deViator ¤
http://deviator.si ¤
http://skylined.org ¤