In live streaming a continuous source — a camera, a switcher, a screen capture — feeds the encoder while viewers are already watching. The defining constraint is the absence of a finished file. The packager publishes new segments every two, four or six seconds, and the manifest grows or rewinds as a sliding window. The player must keep up: choose a starting position close to the live edge, drift toward it when the buffer grows, hold back when the network stutters.
Live splits into latency tiers. Classic HLS and DASH ship at 20–45 seconds behind live and are the default for sports and big events on smart TVs. Low-latency HLS and DASH (LL-HLS, LL-DASH) cut this to 2–6 seconds using partial segments and CMAF chunks. WebRTC and Media over QUIC reach 200 ms to 1 second and are needed for auctions, betting, casino games and any second-screen interaction tied to the broadcast. The whole industry trends downward, but cheaper means slower, and most live operators carry two or three pipelines in parallel.
Live streaming costs more per viewer than VOD for two reasons. It demands more origin and packager capacity, since the segments cannot be pre-baked and pushed once. And it puts heavy pressure on CDN cache fill — at the moment a goal is scored, every viewer in a country wants the same segment within the same second, and the edge must already hold it. Designing a live pipeline is mostly designing for that simultaneity.
