The trap
A Dante network can run for a long time on defaults. Devices find each other with link-local (APIPA) addressing, PTP elects a clock leader on its own, and if everything sits on one switch, it mostly works.
Then something moves to another switch, or another building, and the problems start: clock drift, extra latency, devices that come and go in Dante Controller. That’s what I walked into. The full project is Dante audio network remediation.
Start with observations
Before changing anything, write down:
- which devices are having trouble, and which switch each one is on
- the current PTP leader, and whether it’s the device you’d have picked
- the clock source for each device (network PTP or external word clock)
- the exact error text and when it happens
- what addressing the devices are actually using
In my case, the pattern jumped out once it was written down: the healthy devices shared one access switch, and the troubled ones didn’t. The one console that was fine on another path ran on external word clock, so it wasn’t depending on network PTP for timing.
What “leaving the switch” exposes
On one switch, traffic never crosses an uplink, so missing QoS doesn’t show. Across uplinks, PTP packets compete with everything else. Timing suffers first, then audio.
Being on the same switch as a working device is not proof that a device is configured correctly. It may just mean the problem hasn’t been tested yet.
What fixed it
- A dedicated Dante subnet with static addressing
- Dante Domain Manager with a domain per facility
- A dedicated PTP leader instead of whatever wins election
- A pilot group across buildings before the full cutover
- Disabling stray secondary interfaces that were sitting on APIPA in switched mode
What’s still open
QoS. The network still has none end to end, and that’s the next project: Dante QoS for broadcast networks.
Rules I’d keep
- Change one variable at a time and keep before/after evidence.
- Pick the clock leader deliberately.
- Test across switches and buildings, not just on the bench.