Re supposed L4S flow domination
jlivingood opened this issue · comments
The doc suggests L4S flows will dominate non-L4S flows. The Comcast L4S trials have so far not demonstrated this. In fact, there are cable modem config changes being tested now that would fix a bit of under-performance of L4S flows vs classic flows.
This could be due to burst-intolerance in the L queue of the DualPI2 AQM, which can lead to underutilization for flows in L. My two-flow tests were done in a netem environment without simulating network burstiness, and presumably the Comcast tests involve bursty traffic.
I haven't seen anyone do a two-flow test between a CUBIC and a Prague flow in a bursty environment. That might yield results closer to what you're seeing.