Howto speedup render performance

[notdefine]

Hello everyone, and a happy new year,

I use MLT in combination with icecast to stream videos. It works in principle, but the speed on the server is poor. Locally on my laptop, it runs two to four times faster.

I realize that it could simply be due to the server performance, the system should be faster than my own in terms of performance data. But maybe someone else has an idea ;)

LC_NUMERIC=“C.UTF-8” xvfb-run -a melt -silent example.mlt -consumer avformat:icecast://source:[email protected]:8000/video.webm threads=4 f=webm vcodec=“libvpx” acodec=“libvorbis” ab=“112k” vb=512k in=0 out=100 scale=0.3

• I have to set LC_NUMERIC, otherwise the scale parameter will be ignored, because the separator for floating point numbers in Germany is “,”.
• I use xvfb-run because the server does not have a graphical user interface running.

What would be the possibilities to increase the performance? My tests show that the streaming itself is not the problem, if I write directly to the file system, the performance is not good either.

[ddennedy]

To make a usable comparison, you should use the same build on each system because the one on the server might be built for the old, limited i386 instructions and minimal assembly especially for SIMD. And that affects more so FFmpeg and its libraries such as libvpx. You can use our docker image on both to do that. Or, you can use Shotcut portable build (must use the included launch scripts).

Someone once suggested that xvfb-run was making melt run slow, and I did not believe it. But I think if you move the libmltqt6.so out of the way if you are not using it, then maybe you do not need xvfb-run.

[bmatherly]

What is the CPU utilization on the server when the performance is poor? Have you tried changing values for threads and real_time?
https://mltframework.org/plugins/ConsumerAvformat/

Some operations in MLT do not multithread well - and as a result, they perform better on a system with better single core performance even if it has fewer cores.