Live Is Hard
A live stream has one source and possibly millions of viewers, all wanting the same moments with low delay. CDNs deliver this by turning a continuous stream into cacheable pieces.
Segmented Streaming
The encoder cuts the live feed into short segments, each a few seconds long, plus a manifest that lists the latest segments.
- Players fetch the manifest, then download segments over plain HTTP.
- Because segments are ordinary files, edge caches store and reuse them.
- Once one viewer pulls a segment to an edge, all viewers there share it.
This makes live behave like cached static content despite changing constantly.
The Edge Fan Out
- The origin produces each new segment once.
- Edges fetch a segment on first demand and serve it to their whole audience.
- Tiered caching and shields keep origin fetches to roughly one per segment.
A short cache lifetime matches each segment lifespan, since old segments quickly become irrelevant.
The Latency Trade Off
Shorter segments cut glass to glass delay but increase request overhead and weaken caching. Protocols using small chunked transfers push latency lower while keeping the cacheable model. Operators tune segment length to balance delay against scale.
Key idea
Live streaming chops the feed into short cacheable segments listed in a manifest, so edges fetch each segment once and fan it out to many viewers, with segment length trading latency against scale.