LOW LATENCY REALITY CHECK Topics Low Latency Protocols

LOW LATENCY REALITY CHECK

Topics ● Low Latency Protocols ● Trade-offs ● Video Player Challenges

Streaming Protocols Playback support in HTML 5 RTMP WEBSOCK ETS HLS/DASH CMAF CTE LL-HLS Web. RTC NO YES YES YES

Streaming Protocols RTMP WEBSOCK ETS HLS/DASH CMAF CTE LL-HLS Web. RTC NO YES YES YES Latency LOW MEDIUM LOW VERY LOW Allows multi-bitrate playback YES NO YES YES NO Supports DRM NO YES YES NO Full control over start-up logic YES YES MAYBE Quality assured delivery YES YES YES NO Cacheable by HTTP services NO NO YES YES NO Playback support in HTML 5

Streaming Protocols RTMP WEBSOCK ETS HLS/DASH CMAF CTE LL-HLS Web. RTC NO YES YES YES Latency LOW MEDIUM LOW VERY LOW Allows multi-bitrate playback YES NO YES YES NO Supports DRM NO YES YES NO Full control over start-up logic YES MAYBE Quality assured delivery YES YES YES NO CDN Compatible NO NO YES YES NO Playback support in HTML 5

Trade-offs Visual quality ↔ end-to-end latency Buffer size (and network independence) ↔ end-toend latency Join latency (zapping time) ↔ end-to-end latency Switch latency ↔ GOP size (and bandwidth overhead)

“Current OTT video distribution approaches require a trade-off between Latency, Scalability and/or Viewer Experience” This trade-off results in suboptimal video delivery solutions

TO 1: Visual quality ↔ end-to-end latency

TO 2: Buffer size (and network independence) ↔ end-to-end latency

TO 3: Join latency (zapping time) ↔ endto-end latency

TO 4: Switch latency ↔ GOP size (and bandwidth overhead)

Video Player Challenges Bandwidth measurement Timing Synchronization

C 1: Bandwidth measurement • Chunking affects throughput active segments • Longer time-to-download • Affects ABR algorithms Periods of zero throughpu t

C 2: Timing • Clock accuracy on clients • $Number$ Mode manifests • DASH-IF-IOP-v 4. 2 6 methods: UTCTiming element urn: mpeg: dash: utc: http-iso: 2014 • Client aware + Accurate time server

C 3: Synchronization • Client Buffer & Join Latency • Cross Device • Cross Protocol

C 4: i. OS • Native player doesn’t make it easy to optimize for low latency • App Store policy • DRM

C 5: LL-HLS • Low-Latency extension by Apple on the HLS spec • HTTP 2. 0 PUSH • Playlist Variations • Everyone is still figuring it out • Still in draft, Maybe Q 2 -Q 3 of 2020

Conclusion • There is more to Low Latency than meets the eye • This is a game of trade-offs • There are solutions on the client-side that can cater to your case

But wait! There’s more!

HESP: High Efficiency Streaming Protocol ○ Sub-second latency allowing for interactivity ■ Outperforms CMAF-CTE ○ Remove 10 -20% bandwidth overhead for significant cost reduction ○ Cost efficient scaling over HTTP 1. 1 based CDNs ○ User experience: ■ Instant zapping (~60 ms) ■ Playback across all platforms (including i. OS) ■ Adapting to different playback conditions (network, device capabilities, …)

����♂� Questions? ������ thijs. lowette@theoplayer. com willem. desaegher@theoplayer. com Booth #N 469 �� Get Cosy �� http: //bit. ly/35 BIt. Fg