Less than 2% cpu usage with substreams....down from 25%

sorka

Pulling my weight
Aug 9, 2015
368
200
Upgraded from BI 4 to 5 today. I didn't even know about the substream thing. My motivation was Deepstack integration. But then I read the note about substreams. I have 14 cameras ranging from 2 to 5 MPs. I enabled and configured the substreams on each one and had BI rescan the ONVIF cameras and in some cases I had to manually select the cameras. In one case I even had to go hunt down the path for the second stream for a Sony camera.

As I enabled the substream on each camera configuration, the cpu dropped and kept dropping. By the time I was done, my cpu usage was bouncing between 1 and 2%. I couldn't believe it. I had to run around to each camera and make sure motion detection was really working which is just as well as it was before. On top of that, I haven't yet disabled object detection as is often recommended when integrating with Deepstack.
 
IMO, 5.2.7 of May 1, 2020 when substream RTSP support was added is one of the most valuable improvements in BI since v5 came to exist.
 
Upgraded from BI 4 to 5 today. I didn't even know about the substream thing. My motivation was Deepstack integration. But then I read the note about substreams. I have 14 cameras ranging from 2 to 5 MPs. I enabled and configured the substreams on each one and had BI rescan the ONVIF cameras and in some cases I had to manually select the cameras. In one case I even had to go hunt down the path for the second stream for a Sony camera.

As I enabled the substream on each camera configuration, the cpu dropped and kept dropping. By the time I was done, my cpu usage was bouncing between 1 and 2%. I couldn't believe it. I had to run around to each camera and make sure motion detection was really working which is just as well as it was before. On top of that, I haven't yet disabled object detection as is often recommended when integrating with Deepstack.

Are you using the subs for continuous recording? Hence the drop in cpu?

I use the subs for continuous (clones) and mainstream HD only for snapshots when triggered (deepstack/AI).
i7 -7700 3.4ghz (4 core/8 threads)

I think the high-resolution snapshots are more accurate for object detection. please correct me if I'm wrong.
 
  • Like
Reactions: Flintstone61
So far, the consensus is not to use high resolution for DS detection. As with everything YMMV.
 
  • Like
Reactions: Flintstone61