SRT latency is not a quality guarantee: six contribution-feed tests

SRT feeds can stay connected while late retransmissions damage live video. Six tests cover RTT, negotiated latency, packet recovery and decoded quality.

Kyle Suess3 min read

Share

Share on LinkedIn
Archival photograph of a television broadcast studio, photographed in 2009; illustrative production equipment, not an SRT deployment.

Dek: An SRT contribution feed can remain connected while late packets still damage the decoded programme; validate recovery against actual network delay, not the connection indicator.

A green SRT connection light answers a narrow question: the transport session is established. It does not prove that every media packet arrives before its playback deadline. For a live broadcaster, the distinction matters because a feed can look stable in network monitoring while intermittent macroblocking, missing audio or subtitle corruption reaches master control.

SRT uses retransmission and a receiver buffer to recover missing UDP packets. Its latency setting reserves time for those retransmissions. Too little time means a packet can be recovered by the network after it is already too late for live decoding. Too much buffering adds delay that may be unacceptable for a remote production or a tightly timed handoff.

Why it matters

A contribution engineer should judge SRT by timely media delivery, not just link uptime or a headline latency number. Haivision's SRT latency documentation, updated February 13, 2026, describes a starting rule of thumb around four times round-trip time on a relatively clean network, with important caveats for burst loss. AWS's MediaConnect troubleshooting guidance recommends configuring latency no lower than three times peak observed RTT in its workflow.

Those are engineering starting points, not universal service guarantees. A 40 ms average RTT says little about a path that occasionally spikes far higher during congestion.

What the protocol actually negotiates

The SRT socket-option reference explains that the effective receive latency is the maximum of the receiver's configured receive latency and the sender's peer latency. Operators who inspect only one endpoint can therefore misread the actual buffer delay.

The same reference ties the receive buffer requirement to latency multiplied by bitrate. A setting that works for a modest HD contribution may require a different memory and flow-control budget at substantially higher rates. The protocol can be connected, and retransmission counters can be moving, while packets still miss their deadline.

Six acceptance tests before an event

  1. Measure peak RTT, not just the average. Collect a representative sample across normal operation, busy periods and the expected event window. Compare the worst sustained spikes with the negotiated SRT latency. Document the margin, not just a passing average.
  2. Inject controlled burst loss. In a test environment, introduce isolated loss and short bursts while recording recovered and unrecovered packet counts. Compare the result with the decoded audio and video, not just the transport statistics. A retransmission arriving after its delivery deadline does not repair the viewer's picture.
  3. Read both endpoints. Capture sender peer latency, receiver latency and the negotiated effective value. Check that monitoring reflects the negotiated value rather than an operator's requested setting. Record settings with the event configuration so that later changes are auditable.
  4. Validate buffer and bandwidth headroom. Test at the intended peak bitrate, including variable-rate bursts. The SRT project's configuration guidelines explain receiver-buffer and flow-control limits. Watch for congestion or buffer pressure that a simple packet-loss percentage would miss.
  5. Compare transport recovery with decoded impact. Monitor the MPEG-TS services or other payload before and after the SRT hop. Check continuity, audio, captions and visible picture defects with aligned timestamps. A packet-loss alarm is useful; knowing whether it damaged the programme is operationally decisive.
  6. Rehearse failover with impaired sources. Test complete loss, a degraded-but-connected source and recovery of the primary. AWS's May 1, 2023 MediaConnect SRT failover announcement describes source failover, but a source switch should still be verified against the decoded output, service mapping and the operation's own interruption budget.

The engineering decision

The practical trade-off is not simply lower latency versus higher latency. It is the smallest buffering delay that survives the real network's burst behavior while preserving the programme's acceptable end-to-end timing.

For the next live contribution event, ask three questions: What is the worst relevant RTT? How many packets were not recovered before their deadline? What did the decoder actually output when the path deteriorated?

Treat a stable SRT connection as a prerequisite, not proof of broadcast quality. Capture those three answers in the event acceptance report and repeat the test when the network path, bitrate or endpoint software changes.

Image credit: Alex Dingwall, Broadcast Studio, 2009, public-domain photograph. Archival illustrative image, not an SRT deployment or incident. A crop and JPEG recompression are planned for the hero.