Key takeaways
- Multicast is the most efficient ingest inside a managed network, but it depends on correct IGMP behaviour end to end.
- RTMP and RTSP are connection-based: great for contribution and cameras, but every reconnect must be handled gracefully.
- HTTP and HLS pull sources are easy to reach across the internet, at the cost of extra latency.
- Reconnect with back-off, keep timestamps continuous and verify every source before it goes on air.
UDP multicast and IGMP
In headends and operator networks, live channels are commonly carried as MPEG transport streams over UDP multicast. One sender transmits each channel to a multicast group address, and any receiver that joins the group gets a copy. The network replicates packets only where receivers exist, so adding another streaming server does not add load on the source.
Receivers join groups with IGMP (Internet Group Management Protocol). For this to work reliably, switches need IGMP snooping configured consistently and the network needs an IGMP querier, otherwise group memberships can time out and a channel silently stops arriving. Multicast is also designed for managed networks: it rarely crosses the public internet, so remote sites usually receive channels another way.
Some feeds use RTP on top of UDP, which adds sequence numbers and timestamps. Sequence numbers make it possible to detect lost or reordered packets — useful information for monitoring even when no retransmission is available.
Jitter and packet loss
UDP has no retransmission. A lost packet is a missing piece of video, usually visible as blocking or a brief freeze until the next keyframe. Jitter — variation in packet arrival time — is absorbed by buffering on the receiving side, but excessive jitter forces larger buffers and adds delay.
In practice, ingest reliability comes from the network as much as from software: dedicated VLANs for video, enough headroom on uplinks, correctly sized socket buffers on receiving servers, and network interface tuning where traffic is high. When a source is truly unreliable, the answer is redundancy — a second source for the same channel — rather than ever-larger buffers.
RTMP: push and pull
RTMP runs over TCP and has long been the standard for contributing live video from encoders and production software. It works in two directions:
- Push: the encoder connects to the platform and sends the stream. Convenient for remote events and studios behind NAT, because only the platform needs a reachable address.
- Pull: the platform connects to an RTMP server and requests the stream. Useful when the stream already exists elsewhere.
Because RTMP rides on TCP, lost packets are retransmitted rather than dropped, but congestion turns into delay and, eventually, disconnects. The platform's job is to detect a stalled connection quickly and recover without spiking load on the remote side.
RTSP and IP cameras
RTSP is the control protocol used by most IP cameras and many encoders. It negotiates a session and then delivers media over RTP, either on UDP or interleaved on the same TCP connection. TCP interleaving is often the safer choice across firewalls and lossy links.
Cameras are frequently the weakest link: limited processing power, variable frame rates and firmware that does not always handle many simultaneous clients. Pulling one stream into the platform and distributing it from there protects the camera and keeps every viewer on a consistent feed.
HTTP and HLS pull sources
When a channel must cross the internet, HTTP-based sources are the most practical: they pass firewalls, work with CDNs and use TCP. A source may be a continuous HTTP transport stream or an HLS playlist with segments.
HLS sources add latency — at least a few segment durations — and require the receiver to poll the playlist and fetch segments on time. They are ideal for contribution from remote locations or partners, less so where every second of delay counts.
| Protocol | Transport | Typical use | Strengths | Watch out for |
|---|---|---|---|---|
| UDP multicast (MPEG-TS) | UDP, optionally RTP | Headend and managed networks | Efficient, low latency, scales to many receivers | IGMP configuration, no retransmission |
| RTMP | TCP | Encoder and studio contribution | Simple, NAT-friendly push | Delay under congestion, reconnects |
| RTSP | RTP over UDP or TCP | IP cameras, some encoders | Universal on cameras | Weak devices, client limits |
| HTTP / HLS pull | TCP | Remote and partner sources | Crosses firewalls, CDN-friendly | Higher latency, playlist timing |
Reconnects, timestamps and source checks
Reconnect with back-off
Sources fail. What matters is how the platform responds. Reconnecting instantly in a tight loop — multiplied by hundreds of channels — can overload the upstream server, a firewall or the platform itself the moment a shared link flaps. Reconnects should use increasing back-off and some throttling across channels, so that recovery is fast for one channel and orderly for many.
Keep timestamps continuous
Every reconnect, encoder restart or source switch can reset timestamps. Players expect presentation timestamps to move forward smoothly; a sudden jump backwards can freeze playback or break audio sync. A good ingest pipeline regenerates or corrects timestamps where needed, so the output stays continuous even when the input is not.
Verify sources before they go on air
Before a new source is put into service — or added as a backup — check what it actually carries: video codec, resolution, frame rate and audio tracks. A backup source that is offline, or that delivers a different format than the primary, is not a real backup.
How SaurusRex handles it: channels can ingest UDP multicast, RTMP, RTSP and HTTP/HLS sources, with several sources per channel and automatic failover. The streaming engine restarts failed processes with back-off, throttles reconnect storms, can regenerate timestamps per channel, and a "Scan sources" action checks codec, resolution, frame rate and audio of every source — main and backups — before they are needed. Explore the platform.