Please provide complete information as applicable to your setup.
• Hardware Platform (Jetson / GPU): Jetson Orin DevKit
• DeepStream Version: 6.1.1
• JetPack Version (valid for Jetson only): 5.0.2 GA Revision 1
• TensorRT Version: 8.4.1.5
• Issue Type( questions, new requirements, bugs): question
My setup has cameras which run in trigger mode (i.e., a camera takes image if it’s triggered, otherwise, no image is taken). I notice that if the trigger is activated rapidly, then the total latency is around 40ms. However, whenever the trigger is idle for a about 10 seconds, then it got triggered, the first buffer will have a significant increase in total latency of about 100ms, then subsequent buffers latency drop back down to 40ms (these subsequent buffers must arrive very close after one another for this drop in latency to happen and the total latency of 40ms to be maintained).
I collected latency for each plugins in my pipeline and it turned out that the nvvideoconvert plugin caused the significant increase of latency after an idle period (i.e., no images taken by the cameras for a few seconds).
I collect 3 samples, here are the plots:
As you can see, the jumps in total latency correspond to the jumps in nvvideoconvert latency.
Why does this phenomenon happen? How do I prevent it?