This article is part 1.2 of the "21 Thoughts on Video Streaming in 2021"-series.
Segmented media end-to-end latency will dip below 5s, buoyed by production implementations of LL-HLS and LL-DASH.
HLS and DASH are the two popular types of adaptive streams. (You know, those streams where the video quality is good if you're on good internet, and bad if you're on bad internet.)
LL-HLS and LL-DASH are extensions to HLS and DASH. More specifically, these extensions decrease the "latency" of a live stream. For example, instead of the video being 40s behind the actual live point, you might only be 5s behind the live point with LL-HLS and LL-DASH.
- LL-DASH has been around for a while, and uses techniques like CMAF-CTE (Chunked Transfer Encoding) to reduce the latency.
- LL-HLS wasn't officially supported by Apple until September 2020. The industry is still trying to implement the LL-HLS extension. (It's still quite uncommon to see LL-HLS stream on non-Apple hardware.)
The Wowza "Streaming Latency and Interactivity Continuum" below is one of the most popular graphics in the industry to visualize the latency of the different streaming protocols, including HLS / DASH versus LL-HLS / LL-DASH.
While regular HLS and DASH have latencies of 18s+, LL-HLS and LL-DASH allows you to get below 5 seconds.
Playing an Apple LL-HLS stream in Safari shows that latencies of 5 seconds with LL-HLS are possible, as demonstrated by the screenshot below.
Playing an Akamai LL-DASH stream in dash.js shows that latencies of 5 seconds with LL-DASH are possible, as demonstrated by the screenshot below.
I kinda agree with the prediction that latency will get below 5 seconds. But, I don't think 2021 will be the year where most vendors offers LL-HLS and LL-DASH.
- People are still working on LL-HLS. Doing LL-HLS in combination with DRM isn't straightforward, neither is cross-platform playback, nor is SSAI.
- Doing all of this on LL-DASH is also still pretty hard! (But it's in a more mature state.)
But, let's be happy. Getting below 5s of end-to-end latency is possible in 2021, and that is darn exciting.
(💬 Share your take on low latency through LinkedIn!)