A Little (video) Goes a Long Way
summary
Showmax switches to next‑gen codecs and rewrites the encoding pipeline so a single 90‑minute movie goes from taking multiple hours (often up to 3.5× video duration) to as low as 0.12× duration by splitting mezzanine files into chunks and running parallel chunk workers that transcode and reassemble. Uses ffmpeg, a CMS to set priorities, and Unified Streaming filters; outcome is ~30% less delivered data, far faster job prioritization, and finer device‑codec control.
AI-written summary · about
chapters (23)
- 0:00Intro
- 0:38What is Showmax?
- 0:52Why we care about video size
- 2:14Africa is...not fast
- 4:51Shrink Videos possibilities
- 5:35Capped CRF
- 7:11Next-gen Codecs
- 9:28Why keep using H264?
- 10:27Initial Tech Stack
- 11:14Manifest Example Smoothstreaming
- 11:32Initial State - Pipeline
- 12:40Initial Flaws
- 14:12Parallel Tech stack
- 14:30Parallel Solution
- 16:16Gotta go fast!!!
- 17:12More control over your pipeline
- 17:50Parallel Advantages
- 19:31Clients part
- 20:22Capability detection - Android
- 21:34Editing Manifests - HLS
- 22:24USP Filter
- 23:39Question Time!
- 29:50Conclusion
related talks
26mOptimizing Video Performance for Web Developers
20mStreaming at 250 kbps: raise your expectations
24mPeople make the difference, or video streaming
22mProviding better video experiences for the next billion users.
50mDemuxed 2023 Lightning Talks
17mDelivering better video encodes for legacy devices
19mLow Latency Live from a Different Vantage Point
41m8K Codec Low-Latency Livestreaming, and AI Streaming