Remix VOD

Overview

Remix VOD enables server side mixing of content to produce a single output stream. Clients and playback devices see the stream coming from a single origin and having a timeline without any discontinuities.

The solution makes use of two Unified Streaming Platform products:

  • Remix, to create a single presentation from multiple input sources
  • Origin, for Just-In-Time packaging/transmuxing to all supported streaming formats

Remix VOD allows for the following use cases:

Components

Unified Origin

Playout Origin for content, this takes the remixed MP4 created by Remix and plays out all streaming formats, optionally with DRM applied.

Remix Origin

Based on a SMIL playlist Remix compares the listed sources and produces a single remixed MP4.

Remix Caching Proxy

The Remix process is relatively intensive, but only really needs to be done once for any given presentation. The caching proxy sits between the Unified Origin and Remix Origin to prevent unnecessary repetition of the Remix process.

This can be any HTTP caching proxy, e.g. Apache httpd with mod_cache and mod_proxy, NGINX, Varnish.

SMIL Origin

Creates a playlist based on a request URL.

This is the most important integration point as it provides the mapping of a request URL to the source media that will be used to produce the final presentation.

Integration

SMIL Origin

The SMIL Origin must return a SMIL 2.0 (Synchronized Multimedia Integration Language) playlist.

SMIL functionality supported by Remix

  • 'outputDescription' meta element in head to point to target profile
    • Without setting this, the first clip in the playlist will be used for the target profile
  • 'seq', sequence of clips to playout
  • 'par' combine multiple files into a single clip as part of the sequence, e.g. if different bitrates are stored in separate source files
  • 'clipBegin', 'clipEnd', to select only parts of the clip rather than whole
    • currently only supports using the "wallclock(ISO 8601)" format, not all formats included in SMIL 2.0

SMIL Examples

A SMIL playlist for a simple use case that will play a pre-roll bumper and 30 seconds of Tears of Steel looks like this:

<?xml version="1.0" encoding="utf-8"?>
<smil xmlns="http://www.w3.org/2001/SMIL20/Language">
  <head/>
  <body>
    <seq>
      <video src="http://sample-content/logo_5s_dref.mp4"/>
      <video src="http://sample-content/tears-of-steel-dref.mp4" clipEnd="wallclock(1970-01-01T00:00:30.000Z)"/>
    </seq>
  </body>
</smil>

An example for a use case with pre- and mid-roll advertisements using Sintel as the main content, targetting Sintel for the output profile:

<?xml version='1.0' encoding='UTF-8'?>
<smil
  xmlns="http://www.w3.org/2001/SMIL20/Language">
  <head>
    <meta name="outputDescription" content="http://storage/main/sintel/sintel_dref.mp4"/>
  </head>
  <body>
    <seq>
      <video src="http://storage/ads/origin/origin08_x264.mp4"/>
      <video src="http://storage/main/sintel/sintel_dref.mp4" clipEnd="wallclock(1970-01-01T00:00:30.000Z)"/>
      <video src="http://storage/ads/capture/capture10_x264.mp4"/>
      <video src="http://storage/main/sintel/sintel_dref.mp4" clipBegin="wallclock(1970-01-01T00:00:30.000Z)" clipEnd="wallclock(1970-01-01T00:01:00.000Z)"/>
    </seq>
  </body>
</smil>

An example based on different bitrates in separate MP4s, using <par> to combine these MP4s into a single item:

<?xml version="1.0" encoding="utf-8"?>
<smil xmlns="http://www.w3.org/2001/SMIL20/Language">
  <head>
  </head>
  <body>
    <seq>
      <par>
        <video src="http://storage/path/to/files/file_1280.mp4" />
        <video src="http://storage/path/to/files/file_1024.mp4" />
        <video src="http://storage/path/to/files/file_768.mp4" />
        <video src="http://storage/path/to/files/file_480.mp4" />
      </par>
      <video src="http://local-storage.unified-streaming.com/demo/tears-of-steel/tears-of-steel-teaser-no-jpg.ism" />
    </seq>
  </body>
</smil>

Storage

Storage for Remix VOD can consist of any form of attached storage (a mount point on the filesystem) or HTTP accessible storage such as S3.

The only requirement is that content must be accessible at the same path from both Remix Origin and Unified Origin, e.g. if using network attached storage mounted to /storage, the mount must be the same on all Remix Origins and Unified Origins.

Content Preparation

Input files for Unified Remix should be ISO BMFF / MP4s, and both fragmented or progressive formats are supported.

For optimal performance a data reference MP4 can be made to minimise the index lookups required by Remix. See --use_dref_no_subs

For example, creating a dref MP4 for Tears of Steel with different bitrates in separate files:

#!/bin/sh
mp4split \
  --license_key=<license_key> \
  --use_dref_no_subs \
  -o tears_of_steel_dref.mp4 \
  tears_of_steel_4000k.mp4 \
  tears_of_steel_3000k.mp4 \
  tears_of_steel_2000k.mp4 \
  tears_of_steel_1200k.mp4 \
  tears_of_steel_600k.mp4 \
  tears_of_steel_eng.mp4

Deployment

Remix VOD is flexible when it comes to deployment options, as both Unified Origin and Remix are stateless and work over HTTP. This means it can easily be deployed on both physical or virtual hosts, or using container technology such as Docker.

Our GitHub Remix Demo has each component running in a separate Docker container, and uses Docker Compose to deploy and configure them to produce a working environment.

Supported OS and software versions

Remix requires Apache 2.4.x.

Recommended OS and web server versions are:

Ubuntu 14.04 Apache/2.4.7
Ubuntu 16.04 Apache/2.4.18
Alpine Linux v3.4 Apache/2.4.27

Environments

Remix VOD can be deployed using various approaches:

  • Bare metal
  • Virtual machines as for instance Amazon EC2
  • Container environments managed by for instance Kubernetes

Requirements

  • Content encoding profiles should match
    • Remix uses a set of heuristics to make "best fit" matches of tracks between separate source media
    • Video aspect ratio changes can cause playback quality issues depending on the player as it may stretch the content to fit the expected aspect ratio
    • Some audio differences will cause playback to break and so cannot be matched, for example,
      • sample rate mismatch
      • channel count difference