How the buffer works
A player downloads video slightly ahead of what is being shown and keeps it in a buffer, typically 10 to 30 seconds. As long as video arrives at least as fast as it is played, the buffer stays full. When the download falls behind, the buffer drains, and when it is empty playback stops.
Causes on the service side
- A bitrate ladder with no rung low enough for weak connections
- A lowest rung that is still too heavy, or gaps between rungs that are too wide
- A CDN that is slow or congested in the viewer's region
- Poor cache hit rates, so segments are fetched from a distant origin
- Player adaptation that reacts too slowly to a falling connection
- Segments that are too long, so one slow download empties the buffer
- Overloaded origin or packaging during a live peak
Why it matters
Viewers tolerate lower resolution far better than they tolerate stopping. Each stall raises the chance that the viewer abandons the session, and repeated stalls are a common reason for cancelling. Studies of viewing behavior consistently find that even a small rebuffering ratio shortens watch time.
How to reduce it
- Add a very low rung, around 200 to 400 kbps, as a safety net
- Start playback on a modest rung and step up
- Measure rebuffering by region, network and device to find where it concentrates
- Check CDN performance in the regions where your audience actually is
- Tune the player to favor continuity over resolution
- Offer downloads for viewers on unreliable mobile networks
Last reviewed