Streaming & media
Reducing Live Stream Latency: HLS, LL-HLS and SRT Ingest Explained
Understand where live video delay comes from, how LL-HLS changes delivery, and where SRT belongs in the broadcast chain.
Start with the whole signal path
A viewer’s delay includes capture, encoding, contribution transport, transcoding, packaging, distribution and playback buffering. Reducing one buffer may have little effect if another stage holds several seconds of video. Measure glass-to-glass latency with a visible time reference at the source and at the player.
SRT normally carries the contribution feed from an encoder to the ingest platform. HLS and LL-HLS normally deliver packaged video through HTTP infrastructure to viewers. A production chain can use SRT ingest and LL-HLS delivery together.
Compare delivery and ingest separately
The figures below describe common design expectations and targets, not Gloman service guarantees. Actual results depend on the complete chain and the receiving devices.
| Protocol | Main role | Latency interpretation |
|---|---|---|
| HLS | HTTP delivery to viewers | 15–30 seconds is a common conventional deployment range, not a protocol requirement. |
| LL-HLS | HTTP delivery with partial segments | Below 3 seconds can be a target for a tuned end-to-end chain; verify under load. |
| SRT | Encoder-to-ingest contribution | A configured recovery buffer covers network variation; it is only part of glass-to-glass delay. |
Why conventional HLS accumulates delay
HLS publishes playlists and media segments that players retrieve over HTTP. Segment production and playback buffering both contribute to delay. This provides room to absorb network variation and fits CDN caching well, which can be a sensible trade-off for a large audience.
Shortening segment duration alone does not create a complete low-latency service. Inspect encoder GOP structure, packaging delay, playlist freshness and the player’s live offset. Test startup time and rebuffering as well as the delay of an already-playing stream.
What LL-HLS changes
LL-HLS makes partial segments available before a full segment is complete. Its extensions include blocking playlist reloads and preload hints. The packager, CDN and player must support the required behaviour together; a stale playlist or incompatible cache policy can erase the benefit.
Use realistic mobile and fixed-network tests. Reducing the playback safety margin can make unstable connections more visible. Define the acceptable trade-off between delay, picture quality and rebuffering for the event.
Use SRT to protect the contribution feed
SRT uses UDP with packet recovery and a configurable latency buffer. Allow enough time and bandwidth for retransmissions on the contribution path. A buffer that is too small for the path’s round-trip time and loss pattern can leave missing packets at the decoder.
Agree caller and listener roles, firewall access and encryption settings between endpoints. Monitor loss, retransmissions and buffer behaviour during rehearsal. SRT improves transport resilience within its recovery budget; it cannot recover a feed through an indefinitely failed link.
Validate the event before going live
Gloman carries over 70% of Greece’s online radio and TV streaming traffic. For a new workflow, that operating experience should translate into a measured acceptance test, with supported encoders, formats and players agreed for the project.
- Measure each processing stage and the final viewer delay.
- Test adaptive bitrate switching, mobile packet loss and player recovery after an interruption.
- Rehearse the backup contribution feed and confirm how players behave during a switch.
- Monitor playlist age, segment errors, startup time and rebuffering throughout the event.