Hello community! I have RTSP streams from cameras, and I'm interested to save it to a disk and share as MP4. The main problems are the trade off between disk usage and CPU usage without converting the format nor scaling the image sizes.
I'm using Ubuntu 18.04 and saving chunk videos directly from RTSP (h264), resizing the stream without using much more CPU. I'm using ffmpeg and its dependencies installed directly from repositories. Actual test: * I'm using it on Azure, with an Intel Xeon server (for testing purposes I have instanced two cores: https://pastebin.com/VauHZtLx), but the idea is to select a good trade off between performance and cost to get about 70 streams. I have seen that libx264 outputs a generic flags usage trough ffmpeg: * * using cpu capabilities: MMX2 SSE2Fast SSSE3 SSE4.2 AVX FMA3 BMI2 AVX2 But I have the same outputs and CPU times on my I7 4th generation CPU. Using the time command, to get * The input is (full ffprobe on https://pastebin.com/0dT9ttLG): Stream #0:0: Video: mpeg4 (Simple Profile), yuv420p, 640x480 [SAR 1:1 DAR 4:3], 30 tbr, 30k tbn, 1k tbc Stream #0:1: Audio: amr_nb, 8000 Hz, mono, flt * With this CPU I have tested to convert 7~10 streams from 30 to 5 fps and it goes around 100 % without transcoding the input (only downsizing the bitrate and the FPS). I'm testing with the command: time ffmpeg -re -rtsp_transport tcp -i 'rtsp://something:something@fdsfsdfdf' -t 10 -c:v libx264 -b:v 100k -maxrate 100k -bufsize 50k -r 5 output2.mp4 Time output (full output on https://pastebin.com/VTabvgQS): real 0m19.527s user 0m2.510s sys 0m0.102s Problems to solve: * Can I improve the performance of ffmpeg by compiling it? * They are some guidelines to improve performance? Thanks for all! _______________________________________________ ffmpeg-user mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-user To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
