Recording

Catch-up and time-shift: recording architecture for live channels

What it takes to let subscribers pause, rewind and replay live television — from segmenting and retention to disk planning, programme guides and audio/video sync.

Key takeaways

  • Catch-up and time-shift are the same recording infrastructure seen from two angles: replaying past programmes and moving back within the current one.
  • Storage scales linearly: bitrate × time × channels. Plan retention per channel, not globally.
  • Keep the recorder independent from live delivery so a recording problem never interrupts viewers.
  • Clean timestamps and a predictable audio format are what make seeking feel instant and stay in sync.

Catch-up, time-shift and network PVR

The three terms are often used interchangeably, but they describe different viewer experiences built on the same idea — the operator records channels centrally and serves the recording on demand.

  • Catch-up lets a viewer open a programme that has already finished, usually from the programme guide, within a retention window such as 3, 7 or 14 days.
  • Time-shift lets a viewer pause, rewind or restart the programme that is on air right now, and then return to live.
  • Network PVR (nPVR) adds a personal element: the viewer chooses what to keep, and the operator stores or references it on their behalf. Rights agreements frequently treat nPVR differently from catch-up, so it is worth separating the two in contracts and in the platform.

Technically, catch-up and time-shift can share one recording. Whether a viewer may use it, and for how long, is a question of content rights — which is why retention should be configurable per channel.

Segmenting the recording

Recording a channel as one ever-growing file is fragile and makes deletion expensive. Instead, recorders write the stream as a series of fixed-length segments — typically several minutes each, in MPEG transport stream format. Segmenting has practical advantages:

  • Retention becomes simple: expiring content means deleting the oldest segment files, with no rewriting.
  • Failures stay local: a broken segment affects a few minutes, not the whole day.
  • Seeking is fast: the server maps a wall-clock time to a segment and an offset inside it.

Most recorders copy the video stream without re-encoding. That keeps CPU usage low and preserves the original picture quality, so a recording server is usually limited by disk throughput and capacity rather than processing power.

Retention and disk capacity

Storage requirements are easy to estimate because a recorded stream consumes space at its bitrate. One megabit per second equals 0.125 megabytes per second, or about 10.8 GB per day. The table below shows decimal gigabytes for common channel bitrates.

Channel bitratePer hourPer day7 days14 days
4 Mbps (SD / efficient HD)1.8 GB43.2 GB302 GB605 GB
6 Mbps (typical HD)2.7 GB64.8 GB454 GB907 GB
8 Mbps (high-bitrate HD)3.6 GB86.4 GB605 GB1.21 TB

For a line-up of 100 HD channels at 6 Mbps with seven days of retention, that is roughly 45 TB of usable capacity — before filesystem overhead and headroom. Two practical consequences follow:

  • Retention per channel matters. Popular channels may justify 14 days; others may need only 24 hours or none at all.
  • Spread recordings across several disks. Using multiple record paths distributes both capacity and write load, and lets you monitor and alert on each disk individually. A disk that stops responding must never be allowed to stall the rest of the server.

A recorder independent from live delivery

Live viewers should never notice that a channel is being recorded. That means the recording process should run separately from live delivery, so a full disk, a slow array or a crashed recorder has no effect on the live stream.

It is equally important not to open a second connection to the original source just for recording. Contribution feeds and upstream providers often limit concurrent connections, and two independent pulls can drift apart. Recording from the platform's own internal live output keeps one source connection per channel and guarantees that the recording matches what viewers saw.

Large deployments move recording to dedicated archive servers with high-capacity disks, leaving streaming nodes free for real-time work. When a channel's recording moves from one server to another, the platform needs to remember where older segments live so that playback continues across the gap.

The programme guide as an index

Subscribers don't think in timestamps; they think in programmes. The electronic programme guide (EPG) — usually delivered as XMLTV — turns the archive into a browsable catalogue: each programme's start and stop time points to a range of recorded segments.

Two details make or break this experience:

  • Time zones. Guide data, the recording server's clock and the viewer's device may all be in different zones. Store recordings against an unambiguous clock and convert at the edges, or programmes will start an hour early or late after a daylight-saving change.
  • Guide history. The guide must keep past entries at least as long as the longest recording retention, otherwise the archive exists on disk but disappears from the apps.

Seeking, timestamps and A/V sync

Live playback is forgiving: a player that starts at the newest data and keeps going can absorb small irregularities. Seeking is not. When a viewer jumps to a position in the archive, the player starts from an arbitrary keyframe and relies on presentation timestamps (PTS) to align audio and video.

Some sources carry audio with irregular or non-monotonic timestamps — older MPEG-1 Layer II or MP3 audio tracks are a common example. During live viewing this may go unnoticed, but after rewinding, audio and video drift apart or playback freezes. A robust approach is to re-encode only the audio of recordings to AAC with timestamp resampling, while copying the video untouched. The cost is small, because audio encoding is cheap compared with video, and the result is an archive that seeks cleanly on every player.

How SaurusRex handles it: a dedicated recorder runs independently of live delivery and writes channels in 10-minute MPEG-TS segments taken from the internal live stream, with retention set per channel. Recordings can live on dedicated archive servers with multiple disks, each monitored with alerts, and catch-up and time-shift use time-zone-accurate XMLTV guide data. Recorded audio is stored as AAC by default to keep seeking in sync. Explore the platform.

Let's talk

See SaurusRex running on your own network.

Tell us about your channels, sites and subscribers. We'll walk you through a live demo and propose an architecture that fits.