Help setting up over 200 cameras

[Calin125]

Hi,

I was wondering if you could help us with configuring over 200 cameras for our surveillance team. We are currently running 11 docker containers on a bare metal server with an Nvidia Tesla V100S. Each container is responsible for around 16-25 streams. I have configured ffmpeg to utilize the Nvidia GPU. However, the CPU usage is still high. We have built the app from source as we had to customize a couple of things. Also, is there any chance you could share the source code for the frigate:stable-tensorrt image?

Here's a sample configuration I'm using for each instance. Each instance is running on a separate port i.e frigate 1 on port "8971" frigrate 2 on port "8972" and so on... I'm not sure if this is the optimal setup for our use case and hence if you could guide us on how to set it up. Thanks in advance.

config.yml:

docker-compose.yml

[NickM-27]

Is there a reason why you have transcoding configured for every camera to h264? Also that transcoding is configured to be done on CPU not GPU

[NickM-27]

If you don't have some reason to transcode to h264 then yes I'd suggest removing the line. After that we will need to see screenshots of the system and camera metrics so we can know what is using CPU

[Calin125]

Here's a screenshot of nvidia-smi. Can you please let me know what system and camera metrics you would like to see. I'm not really sure how to get them. Can you please guide me on how to get them for you?

[NickM-27]

Settings Cog in bottom left -> System Metrics and then a screenshot of that as well as the camera tab

[Calin125]

Here's a screenshot of the system metrics. Also, notice CPU usage is 56% compared to only 11% for the GPU. However, as I mentioned before, the server is running 11 containers and each container is running one app instance with 16-25 cameras each.

[NickM-27]

Need to see the camera tab as well

[Calin125]

Sorry, I'm not sure what do you mean by sharing the camera tab? We have 16 live streams on there. I cannot share this atm. Do you mean something else? Many thanks

…

On Mon, 23 Dec 2024 at 16:38, Nicolas Mowen ***@***.***> wrote: Need to see the camera tab as well — Reply to this email directly, view it on GitHub <#15650 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUSOEGBC5JJSEKBCJDUV7UL2HAVAVAVCNFSM6AAAAABUC2LX4OVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTCNRVGEZDKNQ> . You are receiving this because you authored the thread.Message ID: ***@***.*** com>

[NickM-27]

Sure, but if you are decoding a high resolution stream and then resizing it to a low resolution this will use more CPU and is a waste, hence why I am asking.

[Calin125]

You're right, makes sense. I'll ask around and see if we can get the low resolutions links. Is there anything else, in your opinion we should do?

[NickM-27]

Not right now, most things seem about as optimal as they could be

[Calin125]

Ok, thank you for the feedback. Much appreciated. Just to confirm, I have removed the following lines in the go2rts streams:
ffmpeg:ar_investment_16#video=H264

Should I also remove the "h264" part from this line: hwaccel_args: preset-nvidia-h264

[NickM-27]

you can if you want to, but it works the same regardless