S02E09: Game of Streams (Game of low-latency streams)
Our Media Solution Consultant, Magnus Svensson, is sharing his reflections from the online streaming industry in this post. This is part of a monthly series so make sure to follow us here if you don’t want to miss an episode.
IBC has just ended and one of the first reflections is that the companies that were hosted in the “temporary” tent that was created a couple of years ago now are big companies, and streaming is a common distribution form for video. The trends observed include scalability, machine learning, personalization, and low-latency. To build a streaming solution is complex and it was clear that scalability and stability were more important than new features. At the same time, you see more and more ways to use the advantage that comes with the new technology and this is where we see machine learning and personalization.
Low latency
FIFA World Cup 2018 marked a milestone when it comes to live streaming. We read headlines every day during the championship with new streaming and online records. Conviva reported that there were in total 393 million streams and Akamai reported 2.7 times more data streamed in Russia than in Brazil in 2014 and 2.3 times than the Olympics in Brazil 2016.
But the event this summer was also the event when the latency that comes with streaming was discovered by the wide audience. While watching a game you could have it spoiled by someone in your surroundings watching with a smaller delay, or through social media. BBC even issued a warning that their BBC iPlayer was up to two minutes behind the actual coverage.
According to the Bitmovin Video Developer Report 2018 (456 survey submissions from over 67 countries), 55% of the global respondents reported that latency is the biggest problem with video technology today.
But is this not a general problem for streaming, for video in demand and other non-live streams, latency is not a factor. Low latency is important for live sports streaming, especially if betting is involved. It has become a top priority for many streaming media broadcasters and content providers.
As latency reduction comes with one or more drawbacks such as scalability and vulnerability to network disturbances you need to carefully consider when and how to reduce latency. The lower latency you go for, the fewer options exist in the market.
What causes the delay?
Every step of the streaming solution introduces some latency, and you must keep latency in mind at every step when designing your system. One of the contributing factors is a side effect of legacy HTTP technology, which was not designed for live streaming. The HTTP protocol has been designed to transport web pages through the internet efficiently in a best-effort mode.
HTTP was not meant to be a real-time protocol, but it was the best choice for streaming video as you could leverage the infrastructure that was already built. When streaming over the public internet has become one of the major distribution methods for video, the choice of HTTP for live events must be revisited.
HTTP based streaming protocols, such as HLS and MPEG-DASH, use a technology where the video is packaged into segments before it’s sent over the internet. A segment cannot be sent until all frames are collected and packaged, and the client usually buffers 2–3 segments before playback. The longer segments you use, the longer delay is introduced.
The technology introduced by Common Media Application Format (CMAF) chunks and HTTP chunked transfer encoding is seen as one solution where you could keep HTTP. Chunked transfer encoding allows a server to maintain an HTTP persistent connection and can be used in combination with MPEG-DASH and CMAF chunks and could also be used by HLS.
Alternative solutions
The alternative to HTTP based streaming is continuous streaming protocol, where you don’t rely on TCP and you don’t package the video into segments and packets, or frames, are sent as soon as they are created.
RTMP and Flash have traditionally been used for live streaming. And even if Flash is dead for streaming, RTMP is still being used. However, RTMP comes with a number of drawbacks such as security issues, reduced support from the CDN vendors, difficulties with firewalls, lack of support for new codecs and ad-insertion support.
SRT (Secure Reliable Transport) is another continuous streaming protocol and is an open source video transport protocol driven by the SRT Alliance. In contrast to RTMP, SRT handles firewall traversal, packet loss, and fluctuating bandwidth. But at current state SRT is more suited for low-latency transmission than to large scale distribution.gra
WebRTC is a technique that is getting more and more implementations for low latency streaming. The protocol was initiated by Google as an open-source solution for browser-based real-time communication. It’s UDP based without the need for media segments to allow for low latency. Limelight just announced support for WebRTC with global scale.
In addition to the standard or open source alternatives, we have proprietary solutions such as Sye from Net Insight and WOWZ from Wowza that delivers ultra-low latency with scale.
What solution to choose from the above alternatives is depending on your use case and targeted reach. You have to evaluate your needs and the pros and cons for each of the options. When HTTP based protocols are the natural choice for most streaming use cases, others could be the best for some.
To watch out for the coming months…
The speakers have been announced for Streaming Tech Sweden, that will take place on the 7thof November in Stockholm. You will be able to listen to speakers from Netflix, YouTube, Akamai and many more and this is a very good opportunity to be inspired and educated by leading technology experts in the streaming industry.
Magnus Svensson is a Media Solution Consultant and partner at Eyevinn Technology. Eyevinn Technology is the leading independent consultant firm specializing in video technology and media distribution.
Follow me on Twitter (@svensson00) and LinkedIn for regular updates and news.
