I have a pipeline decoding 8 IP cameras’ h264 video stream of 1920x1080@25fps simultaneously, and get RGBA video frames via nvvideosink’s EGL stream. Each camera pipeline like this: rtspsrc ! decodebin ! nvvidconv ! nvvideosink.
After profiling with nvvp tool, I find that gst_nvvconv_transform sometimes very time consumed. Normally, it takes within 1ms, but sometimes it takes tens of milliseconds.
By examining the tegrastat’s report, the usages of CPU/GPU/EMC/NVDEC all are not overload.
RAM 4498/7851MB (lfb 5x4MB) cpu [42%@498,off,off,32%@499,40%@499,34%@499] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 13%@522
RAM 4498/7851MB (lfb 5x4MB) cpu [45%@345,off,off,32%@345,34%@348,35%@348] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 31%@420
RAM 4500/7851MB (lfb 5x4MB) cpu [40%@805,off,off,39%@806,48%@807,35%@808] EMC 48%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 13%@318
RAM 4500/7851MB (lfb 5x4MB) cpu [41%@652,off,off,35%@652,36%@653,39%@652] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 74%@522
RAM 4500/7851MB (lfb 5x4MB) cpu [41%@345,off,off,36%@345,35%@344,31%@345] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 46%@420
RAM 4500/7851MB (lfb 5x4MB) cpu [44%@345,off,off,31%@345,37%@345,36%@345] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 51%@420
RAM 4498/7851MB (lfb 5x4MB) cpu [54%@2031,off,off,33%@2035,35%@2030,34%@2034] EMC 48%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 86%@726
RAM 4499/7851MB (lfb 5x4MB) cpu [46%@807,off,off,35%@807,34%@806,32%@807] EMC 48%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 80%@420
RAM 4499/7851MB (lfb 5x4MB) cpu [42%@1420,off,off,35%@1421,31%@1419,36%@1420] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 83%@216
RAM 4499/7851MB (lfb 5x4MB) cpu [38%@806,off,off,40%@806,32%@806,34%@808] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 41%@522
RAM 4498/7851MB (lfb 5x4MB) cpu [43%@960,off,off,33%@960,37%@960,34%@964] EMC 49%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 3%@624
RAM 4498/7851MB (lfb 5x4MB) cpu [43%@653,off,off,32%@652,47%@652,35%@656] EMC 48%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 88%@318
RAM 4498/7851MB (lfb 5x4MB) cpu [40%@805,off,off,31%@806,35%@652,38%@653] EMC 48%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 65%@624
RAM 4499/7851MB (lfb 5x4MB) cpu [43%@499,off,off,33%@499,35%@499,37%@499] EMC 47%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 62%@420
RAM 4464/7851MB (lfb 5x4MB) cpu [47%@806,off,off,31%@806,37%@806,34%@806] EMC 49%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 10%@522
RAM 4500/7851MB (lfb 5x4MB) cpu [54%@653,off,off,36%@652,48%@652,38%@652] EMC 48%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 76%@420
RAM 4500/7851MB (lfb 5x4MB) cpu [45%@345,off,off,35%@345,34%@345,36%@345] EMC 49%@1866 APE 150 NVDEC 576 MSENC 217 GR3D 59%@522
I try running jetson_clocks.sh to maximize the system clock, and the issue disappears. I was wondering is there a problem with the system scheduling. How can i fix this, and any suggestions would be appreciate, thanks.
The attachments are the nvvp profile report, and screenshot of the issue’s timeline.
NewSession1.xml (3.01 MB)
