Question about deploying SRS-WHIP-WHEP with stuttered pictures after a few hours of streaming.

[pingpingz]

Hello,
I would like to ask a question with SRS-Whip

I try to deploy two instance of SRS on two VM
and install ffmpeg build of this git
git clone -b feature/rtc-muxer https://github.com/winlinvip/ffmpeg-webrtc.git

I have to set up Multiple SRS because I can scale multiple SRS-WHEP node to multiple servers and can make load balancing

I deploy as like this Diagram

RTMP Feed --> [SRS 1] --FFMPEG via WHEP--> [SRS 2] --> WHEP Player

RTMP Feed Spec

Encoding : h264 (via nvidia nvenc with OBS)
1920 x 1080 @ 60 FPS
bitrate : 8mbps
Profile : baseline
B-frame : 0

SRS1 Configuration

listen              1935;
max_connections     1000;
srs_log_tank        file;
srs_log_file        ./objs/log/srs.log;
daemon              on;
http_api {
    enabled         on;
    listen          1985;
}
http_server {
    enabled         off;
    listen          8080;
    dir             ./objs/nginx/html;
}
vhost __defaultVhost__ {
    tcp_nodelay     on;
    min_latency     on;

    cluster {
        mode            local;
        origin_cluster  on;
    }

    play {
        gop_cache       off;
        queue_length    10;
        mw_latency      350;
    }

    publish {
        mr_latency      350;
    }

    hls {
        enabled         off;
        hls_path        ./objs/nginx/html;
        hls_fragment    10;
        hls_window      60;
    }
}

FFMPEG Command

ffmpeg -hide_banner -loglevel error -re -i rtmp://10.10.10.139/live/Test -c:v copy -c:a libopus -ar 48000 -rematrix_maxval 1.0 -ac 2 -b:a 192k -f whip "http://127.0.0.1:1986/rtc/v1/whip/?app=rtclive&stream=Test"

SRS2 Configuration

listen              19351;
max_connections     1000;
srs_log_tank        file;
pid                 ./objs/srs1.pid;
server_id           srs-cdn01-01;
srs_log_file        ./objs/log/srs1.log;
daemon              on;
http_api {
    enabled         on;
    listen          1986;
}
rtc_server {
    enabled on;
    listen 8001; # UDP port
    candidate $CANDIDATEIP;
}
vhost __defaultVhost__ {
    tcp_nodelay     on;
    min_latency     on;

    play {
        gop_cache       off;
        queue_length    10;
        mw_latency      300;
    }

    publish {
        mr_latency      350;
    }

    rtc {
        enabled     on;
        rtmp_to_rtc off;
        rtc_to_rtmp off;
    }
}

I set the kernel configureation as in https://ossrs.io/lts/en-us/docs/v5/doc/performance#rtc already

# Query the kernel configuration
sysctl net.core.rmem_max
sysctl net.core.rmem_default
sysctl net.core.wmem_max
sysctl net.core.wmem_default

# Set the UDP buffer to 16MB
sysctl net.core.rmem_max=16777216
sysctl net.core.rmem_default=16777216
sysctl net.core.wmem_max=16777216
sysctl net.core.wmem_default=16777216

The Workflow works all smoothly but after about an hour of streaming on the WHEP Player, but after that it will have stuttered of pictures frames. It is not apply to sound layer, the sound seems fine but not the picture.

What can I do to tune the system to run smoother?

[winlinvip]

Why use WHIP for forwarding between different nodes instead of RTMP:

RTMP Feed --> [SRS 1] --FFMPEG via RTMP--> [SRS 2] --> WHEP Player

TRANS_BY_GPT4

[pingpingz]

I once try to publish via RTMP with

rtmp_to_rtc on;

And it is very consuming the CPU.
I think that srs seems to re-transcode the stream with x264 before sending out with WHEP although the stream is on the right codec (h264 baseline with no b-frame), therefore my CPU can run fine with aac -> opus audio transcoding.

So to make sure that SRS will not re-transcode the stream again, I try to use ffmpeg to transport the stream directly.

Is it anything I miss to config the stream to passthru the re-transcoding process?

[winlinvip]

After enabling the rtmp_to_rtc option, audio will be transcoded from Opus to AAC, but video will not be transcoded.

TRANS_BY_GPT4

[pingpingz]

It's turn out that my encoding option on the source is not correctly configured in OBS, my bad and I'm sorry.

OK, Let me try to start over again.

I try to deploy my system as this...

Source --RTMP-> [SRS : Origin] --Cluster--> [SRS : Remote] --> Output (WHEP/RTMP)

In output, the RTMP works fine, but WHEP does not.

Do I have to use ffmpeg to transport stream myself?

[winlinvip]

Please do not attempt without preparation; first, consult the documentation at the following URLs: https://ossrs.io/lts/en-us/faq#webrtc-cluster and https://ossrs.io/lts/en-us/docs/v6/doc/edge.

TRANS_BY_GPT4

[winlinvip]

The Edge(Remote) mode is specifically designed for live streaming applications. Please ensure it is used correctly and refer to the documentation at https://ossrs.io/lts/en-us/docs/v6/doc/edge for guidance.

If you need to forward streams between multiple SRS instances and use WHEP, you can utilize the Forward mechanism, or employ FFmpeg to pull and forward the stream.

TRANS_BY_GPT4