A half measure
S06E09: Game of Streams (A half measure)
Our Media Solution Specialist, Magnus Svensson, is sharing his reflections on the online streaming industry in this post. This is part of a monthly series so make sure to follow us here if you do not want to miss an episode.
Latency has always been a hot topic for video streaming. And when more latency-sensitive content like sports is moving into the streaming services, it becomes even more important. But how low latency do you need?
For many broadcasters and content owners, the target is to reach at least the same latency for sports as in a traditional broadcast. On average, broadcast latency is around 6–8 seconds which would be low enough for most live content. Sometimes you even have an artificial delay introduced to prevent mistakes or unacceptable content from being broadcasted, sometimes known as the “7 seconds delay”.
Latency on par with broadcast will also in most cases reduce the risk of spoilers from Twitter and other social media. But we have use cases that need lower latency, such as live events with interactivity, betting, companion devices (even if device synchronization is more important for this case), and in-stadium experiences.
Different techniques exist today to achieve low latency streaming. The first step is to tune the existing adaptive bitrate video pipeline to lower the end-to-end latency. On top of this, you could introduce the low latency modes of HLS (HTTP Live Streaming) and DASH which will reduce latency even more. For sub-second latency, you will need either one of the proprietary solutions in the market or use WebRTC. All techniques to lower latency come with some level of increased risk for buffering and quality degradation.
On par with broadcast
To reach latency on par with traditional broadcast it is enough to tune the adaptive bitrate video pipeline. Areas to tune include the content ingest upload time, encoding delay, segment sizes, player buffer, and player design. If all parts are optimized it should be possible to use standard adaptive bitrate with latency on par, or below, the broadcast feeds.
The low latency versions of HLS and DASH will on a very simplified and high-level view chop the segments into even smaller chunks and distribute those to the device as soon as available. This will reduce the latency to around 1–2 seconds, but with the drawback of a more complex video pipeline and increased risk for buffering and quality degradation.
But will 1–2 seconds be enough to handle the really latency sensitive scenarios mentioned above? If you have interactivity or an in-stadium experience, 1–2 seconds will be noticeable and not acceptable. And for betting, especially live betting, every second can be counted in money. With that in mind, are the low latency versions of HLS and DASH a half measure that increases the complexity of HTTP adaptive bitrate streaming but still does not reach the target?
WebRTC
WebRTC is a standard that allows the delivery of real-time content to users, with sub-second end-to-end latency. WebRTC support is built into all modern browsers, and it allows for streaming of video, audio, and data. The original focus of WebRTC has been video conferencing but it is getting more and more attention and usage also for streaming of premium content.
Efforts are ongoing to be able to use WebRTC-based distribution in broadcast streaming in a standardized way. Similarly, to how it works for “traditional” HTTP-based streaming where the ingest part is separated from the distribution side, the WebRTC version would be an ingest protocol and multiple WebRTC media servers in an SFU (Selective Forwarding Unit) topology where you have an origin media server and one or many edge media servers in the edges.
The principle of this setup is to use one WebRTC media server as the “origin” which in turn will relay and forward all the media to other WebRTC media servers acting as edges, and the players will establish a connection with these edge media servers where media will be streamed from. This topology makes it possible to scale up to a larger number of viewers.
So, for most of the live streaming, it will be enough to tune the existing pipeline and reach the same levels as the traditional TV distribution levels. And for the cases that need lower latency, use the distributed and scalable WebRTC architecture will be the way forward.
To watch out for the coming months…
Consolidation, mergers, bundles, and partnerships will accelerate, and to reach further, it is by standing on the shoulders of Giants.
Demuxed will take place in San Francisco on October 12th and 13th. Two days filled with interesting tech talks about video technology. And the week after NAB Show New York will be held. I will be attending both, so please reach out if you want to catch up.
Magnus Svensson is a Media Solution Specialist and partner at Eyevinn Technology. Eyevinn Technology is the leading independent consulting company specializing in video technology and media distribution.
Follow me on Twitter (@svensson00) and LinkedIn for regular updates and news.