The result is shown with the gray square marks in Figure 6: no livelock, and much improved peak throughput (the MLFRR increases from under 2000 packets/sec.
This should reduce the per-packet overhead and so increase the MLFRR.
Note that at rates above the MLFRR of the no-feedback system, even with queue-state feedback the system does lose some packets.
The results for the no-feedback kernel show a noisy relationship between input and output rates, above the MLFRR.
His approach improves system behavior by reducing per-packet and per-byte overheads, and thus increases the MLFRR, but does not directly improve overload behavior.