Cascade Probability Control to Mitigate Bufferbloat under Multiple Real-World TCP Stacks

Abstract

Persistently full buffer problem, commonly known as bufferbloat, causes unnecessary additional latency and throughput degradation whenever congestion happens in Internet. Several proposed queue management schemes, with the debloat mission, are almost based on the modification of one-loop feedback control where the instability and bad transient behavior are still big challenges. In this paper, we present a cascade probability control scheme using margin optimal method to address such challenges under different kinds of real-world TCP stacks. Simulation results guarantee the measured round trip time tracking to a low value of delay (e.g., ≈180 ms under TCP Reno, and ≈130 ms under TCP Cubic) and ≈50% delay reduction in comparison to current deployed queue management schemes in network devices.