Open vr000m opened 5 years ago
This requires rethinking, and perhaps more people's input.
@aboba @Orphis We ICE-boxed this issue a while a go, but there has been plenty of SVC work recently. Is SVC stats something that we should revisit? TPAC could be a time to discuss this? Adding the label without having any suggestions yet or knowing if I'll talk about it. Feel free to take it on
@Orphis What are the metrics that would be of interest on a per-SVC layer basis?
Some of the interesting metrics that immediately come to mind: bitrate, resolution and maybe framerate? I'm not sure how framerate could be different from the "simulcast" level stats ; SVC is hard.
TPAC decision: Something like "svc-layer" stats makes sense (slide), but how to identify the layer is probably more complicated than "unsigned long layer" and if we want to describe relationships between layers that requires more discussions. This is ready for PR if we can figure out what to do with the "layer" as part of the PR
@vr000m Will you look into the layers question and make a PR?
perResolutionFramesSent/Received gives the frame counters for each resolution
distributionOfTimeSinceLastFrameSent/Received gives the amount of time between frames, so these counters would correspond to appropriate time values for the frame rates, for example: 15 FPS (66.66ms), 30 FPS (33.33ms) and 60 FPS (16.66ms)