server
server copied to clipboard
CasparCG
Periodic slowdowns (possibly caused by mixers) in frame production
Expected behaviour
(near) perfect frame production, in a steady pace at correct framerate.
Current behaviour
I'm seeing periodic (it might be around every 5 minutes) slowdowns in frame production, looking at the diag window it looks like mix-time spikes (which might be a cause, or an effect) and consumers not getting their frames in time, after which caspar "catches up" to get out the correct amount of frames. I'm running 4 HD channels with 4 NDI consumers attached on a test bench.
Environment
- Current master
- 2.3-dev
- Debian Buster
Screenshots
My first suspicion was that CasparCG is periodically garbage collecting the OGL buffers, but this doesn't seem to be the case, while the mix-time spike looks similar/the same as when GL GC is called via AMCP there is no indication in the log file that it is clearing the buffers at these moments.
While watching GPU (k620) utilization I see occasional spikes to 100%, while averaging 80ish percent, 2.1 was around 40ish percent, so this might be an explanation.
In 2.2 and later the GPU processes double the amount of data compared to 2.1 for interlaced formats due to its internal progressive pipeline.
In 2.2 and later the GPU processes double the amount of data compared to 2.1 for interlaced formats due to its internal progressive pipeline.
Yep it does :P
Hoping there's something left to optimize, we're intending to upgrade soonish to 2.3, but some of our servers are on k620, which would not cope, and swapping GPU's is not something we'd like to do :/