Multiple encoder session for multiple users - best practice?

Question

Multiple encoder session for multiple users - best practice?

Closed this issue 6 years ago · 7 comments

If I am serving multiple clients, should I be creating separate encoder instances per client or just create one and share them across all clients? I am new to this, any help is highly appreciated.

Thanks!

Answer 1 · 2018-12-18T10:49:46.000Z

When I try create multiple encoder sessions I get this error.

Encode error: Failed to create encoder (NvEncoder : m_nvenc.nvEncOpenEncodeSessionEx(&encodeSessionExParams, &hEncoder) returned error 10 at /home/op/NvPipe/src/NvCodec/NvEncoder/NvEncoder.cpp:54

Answer 2 · 2018-12-18T10:59:23.000Z

Hi,

that depends on what exactly you want to serve your individual clients. If everybody is supposed to receive the same stream, then you only need a single encoder instance. Just remember that you'll probably want to force an I-frame whenever a new client has connected.

Regarding your error: How many encode sessions did you create? Depending on your hardware, you might run into a limit on the maximum number of concurrent encode sessions (e.g., 2 on GeForce).

Answer 3 · 2018-12-18T11:22:07.000Z

Hi,

that depends on what exactly you want to serve your individual clients. If everybody is supposed to receive the same stream, then you only need a single encoder instance. Just remember that you'll probably want to force an I-frame whenever a new client has connected.

Everybody will receive different streams. Is there any example/pointer I can refer for the I-frame concept?

Regarding your error: How many encode sessions did you create? Depending on your hardware, you might run into a limit on the maximum number of concurrent encode sessions (e.g., 2 on GeForce).

I just created 2 sessions.
I have GeForce GTX 1080 Ti with 11GB GDDR5X

Answer 4 · 2018-12-18T12:12:23.000Z

Every encode call has a bool forceIFrame parameter.
But if every client is supposed to receive a different stream, then you need separate encoder instances anyway.

I suggest having a separate encode thread for each client connection. So when a new client connects, start a thread, create an encoder in it, and then encode+send in a loop while the client is connected.

Answer 5 · 2018-12-18T12:39:37.000Z

Sounds good, thanks for the quick responses :)

Answer 6 · 2019-03-29T20:11:57.000Z

I've been getting this same error myself.. In my case:
1 - I've separate threads for encoding
2 - Second encoder instance is created without any error
3 - First frame I send in to the encoder gives me the above error
4 - It encodes from second frame BUT the video resolution is set to 1920x1080 (I'm sending in frames of 1152x648 size)

Any ideas?

Thanks!

Answer 7 · 2019-04-03T15:30:07.000Z

I've been getting this same error myself.. In my case:
1 - I've separate threads for encoding
2 - Second encoder instance is created without any error
3 - First frame I send in to the encoder gives me the above error
4 - It encodes from second frame BUT the video resolution is set to 1920x1080 (I'm sending in frames of 1152x648 size)

Any ideas?

Thanks!

Check out the update to Video Codec SDK 9 that I just pushed. It also includes a fix for concurrent encoding, and a "concurrent" example.