How does low‑latency video streaming work and what factors affect it?

27 Eyl 03:15

Exactly, I ran a live Q&A for a film festival last year and hit the same latency wall. The biggest change in the encoding pipeline is cutting the GOP length down to just a few frames (often 1‑2 seconds or less) and disabling B‑frames, so each keyframe can be shipped almost immediately. Most low‑latency encoders have a “ultra‑fast” or “low‑latency” preset that forces constant‑bitrate output and uses hardware‑accelerated profiles (e.g., NVENC with “low‑latency‑high‑performance”). You also want to package the stream as chunked CMAF or fragmented MP4, which lets you push sub‑second segments (usually 200‑250 ms) instead of the traditional 2‑4 second chunks used for VOD. On the transport side, protocols like WebRTC, SRT, or Apple’s LL‑HLS (and the newer CMAF‑based low‑latency DASH) are the go‑to choices because they support small segment sizes and allow the player to keep a minimal buffer (often just one or two chunks). The client buffer can be forced down to ~300 ms, which together with the encoder latency gets you under a second total delay. Network‑wise, the biggest killers are jitter and packet loss – they force the player to request retransmissions or increase its buffer, instantly blowing up latency. High round‑trip time and sudden bandwidth drops also make adaptive bitrate switches jittery, so a stable, low‑latency CDN or a direct SRT tunnel with forward‑error‑correction usually gives the most consistent results. Monitoring RTT, jitter, and loss in real time and throttling the bitrate just enough to stay above the “buffer‑underrun” threshold is key to keeping that sub‑second window tight.