/lhls-simple-live-platform

Very simple low latency live platform prototype

Primary LanguageShell

lhls-simple-live-platform

This readme shows how to build a very simple live streaming platform based on open source tools with a glass to glass latency around 2s.

To learn more about it see the presentation at syd<video> on 2020/06/24:

Block diagram

Block diagram

Installation

  • Launch EC2
    • Type:
      • t2.small (for transmuxing)
      • t2.xlarge (for transcoding from 720p to 720p and 480p)
    • Security groups
      • Open Inbound TCP 1935 for everybody (RTMP)
        • Open Inbound UDP 1935 for everybody (SRT)
      • Open Inbound 9094 for everybody (HTTP)
  • Update the machine sudo yum update -y
  • Install git: sudo yum install git -y
  • (optional) Install tmux: sudo yum install tmux -y
  • Install (compile) ffmpeg
  • Install DejaVu fonts (to do the overlay on transcoding): sudo yum install dejavu-sans-fonts -y
  • Install this repo
cd ~
git clone https://github.com/jordicenzano/lhls-simple-live-platform.git
go get github.com/jordicenzano/go-ts-segmenter
go get github.com/mjneil/go-chunked-streaming-server

Usage (RTMP source)

  • Create a shell to the EC2 machine (tmux recommended), and start webserver in HTTPS
cd ~/go/bin/
./go-chunked-streaming-server
  • Create ANOTHER shell to the EC2 machine (tmux recommended), start RTMP server + segmenter with a multirendion transcoding configuration
cd ~/lhls-simple-live-platform/scripts/
./transcoding-multirendition-rtmp.sh live

  • Open your favorite RTMP client: OBS, Wirecast, Elemental, Wowza Clearcaster, ffmpeg, etc

    • Configure the RTMP URL as: rtmp://[PUBLIC-IP-EC2]:1935/live/stream
    • Recommended short GOP (1s) and (if possible) activate "zerolatency" video encoding mode
    • Start streaming

  • Tested with following players: Safari, Quicktime and ffplay)

    • Use this URL: http://[PUBLIC-IP-EC2]:9094/mrrtmp/playlist.m3u8

  • Example glass to glass latency with this set up: 2.01s Glass to glass latency

Usage (SRT source)

  • Create a shell to the EC2 machine (tmux recommended), and start webserver in HTTPS
cd ~/go/bin/
./go-chunked-streaming-server
  • Create ANOTHER shell to the EC2 machine (tmux recommended), start SRT server + segmenter with a multirendion transcoding configuration
cd ~/lhls-simple-live-platform/scripts/
./transcoding-multirendition-srt.sh live

  • Open your favorite SRT client: OBS, Wowza Clearcaster, ffmpeg, etc

    • Configure the SRT URL as: srt://[PUBLIC-IP-EC2]:1935
    • Recommended short GOP (1s) and (if possible) activate "zerolatency" video encoding mode
    • Start streaming

  • Tested with following players: Safari, Quicktime and ffplay)

    • Use this URL: http://[PUBLIC-IP-EC2]:9094/mrsrt/playlist.m3u8

  • Example glass to glass latency with this set up: 2.54s Glass to glass latency

Notes

  • This is JUST A PROOF OF CONCEPT / PROTOTYPE do not use it in production
    • Example of known problem: If the encoder reconnect for any reason, the live stream finishes
  • There is a lot going on about how to do ABR when you are sending data at "media speed" and NOT at "cable speed". So for multirendtion and chunk transfer you should expect regular players going to the lower BW lane
  • go-chunked-streaming-server does not send max-age headers, so if you want to add a CDN on top of this prototype to scale this platform up you need to set up the expiration manually