Notes on finding out what actually happened during a live, for the engineers who get the complaint.
When a viewer says the live froze, most teams can’t say quickly whether it was their stream, the viewer’s network or the viewer’s device. Comparing that viewer with everyone else watching at the same moment narrows it down, as two real, anonymised rooms show. Four questions to ask your team this week.
The real cases behind it, with the numbers and the method.
One of 19 viewers in a live video room lost 9.2% of her inbound video packets over 49 minutes, which the room’s pooled loss of 1.1% turned into what looked like a mild problem shared by everyone; without her it was 0.17%. Comparing her with the other viewers of the same three streams at the same moments (pooled loss 0.13–0.21% per stream) showed the fault was on her downlink.
Three real incidents from live video rooms, each sorted into stream, network or device by comparing one viewer with everyone else watching at the same moment. The article also covers the client-side data you need to make the same call yourself.