64 STREAMING MEDIA INDUSTRY SOURCEBOOK 2018
offering low-latency viewing). Nanocosmos took a slightly different approach, allowing broadcasters to
connect via WebRTC, but streaming
on the other end using HLS that is
optimized for low latency.
This takes the single-broadcaster
approach to the next level.
As video becomes more interactive, so do the ways of creating the
video itself. Certain events, such as
webinars and news-related video
streams, need the ability to interview people live and broadcast the
interview as a single unit to viewers.
But the broadcaster and his hosts
might not be located in the same
room, so they need to be on a live
video chat that is then mixed and
streamed toward the viewers.
The media server in Figure 3 now isn’t only a media
gateway. It receives real-time media streams from all
the active participants (the broadcaster and the hosts),
manages the conference call between them, and also
mixes all inputs to create a single stream that can be
digested by a video CDN—most likely in RTMP format.
YouNow was probably one of the first to offer such
a user experience. Recently, Facebook Live as well as
Instagram added support for adding a guest to a livestream event.
In the enterprise, it is quite common to host webinars
that have multiple hosts joining from different locations. In recent years, webinars have been shifting from
voice toward video. This has brought with it the need to
enable these types of interview scenarios. At the same
time, it has increased the requirement for lower latencies for viewers, which brings us to the next use case.
3. Low-Latency Live Streaming
In a way, the first two use cases can be seen as a prelude to this one—achieving low-latency live streaming
in a post-Flash world. Since WebRTC was designed
and implemented from the ground up for real-time
communications, it is quite capable of offering low-latency live streaming.
In the past year or two, we’ve seen companies starting to employ WebRTC for such use cases. The vendors solving this technical problem are tackling it
from two different directions:
1. Taking the video conferencing approach
and increasing it to support larger groups via
a selective forwarding unit (SFU), a type of
2. Starting from the existing RTMP streaming
technologies and enhancing them to support
WebRTC as well.
Both directions are valid, and have different technical benefits and challenges. At the end of the day,
both need to tackle the issue of scale: How do you
take a single stream and broadcast it to a large number of viewers? Even if you assume a single media
server can broadcast to thousands of viewers, what
happens when the stream in question is requested by
even more viewers?
The solution is cascading, which is quite similar to
how CDNs work today. Figure 4 (on page 66) shows
such an architecture, where a single broadcaster’s
stream gets cascaded from one media server to another before reaching its destination.
This is what Beam did with an infrastructure of
cascaded SFUs prior to being acquired by Microsoft and rebranded as Mixer. This is how all the gray
zones of the internet that offer auctions, gambling,
and porn are tackling the issue of live streaming
2018 will see more technology companies offering
such solutions and products, as well as a CDN or two
that will offer such commercial services.
4. P2P Large-Scale Streaming
There are times when a few seconds of latency
are fine, but what if you are trying to deal with large
peaks of viewers at a given moment, as in the case of
the Game of Thrones season finale?
The challenge here stems from the fact that hundreds of thousands or even millions of viewers flock
A typical setup for a multi-person video interview that is mixed and streamed to viewers