The quinn BBR code has a lot of WARNING text. I enabled it anyway because bufferbloat means the death of any prioritization and real-time ambitions.
Somebody should validate that BBR is working as expected. I would use qlog/qvis to monitor the congestion metrics and eyeball if everything looks good for a few crappy sessions. I can tell you from experience that it's very easy to screw up pacing and application-limited flows.
Also, we may want to disable the PROBE_RTT phase when application-limited. I think this would have a lot of impact for live media, even more than switching over to BBRv2.
Also, we'll need to verify that Quinn is probing the network every 1/8th of the time like BBR recommends. This would be implemented as a ton of PING packets filling up the congestion window.
The quinn BBR code has a lot of WARNING text. I enabled it anyway because bufferbloat means the death of any prioritization and real-time ambitions.
Somebody should validate that BBR is working as expected. I would use qlog/qvis to monitor the congestion metrics and eyeball if everything looks good for a few crappy sessions. I can tell you from experience that it's very easy to screw up pacing and application-limited flows.
Also, we may want to disable the PROBE_RTT phase when application-limited. I think this would have a lot of impact for live media, even more than switching over to BBRv2.