Random early detection with flow number estimation and queue length feedback control

We evaluate effects of parameters in queue management on performance using 2k factorial designs. Among these systematic and environmental parameters, number of flows and Pmax are two dominant factors affecting RED performance. Then, we propose a novel active queue management scheme, NRED, employing flow number estimation and queue length feedback control, which are motivated from Bloom filter and control theory, respectively. By estimation of number of active flows, NRED is more scalable than FRED which employs per-active-flow accounting. Furthermore, fluctuation of queue length under NRED is smaller than BLUE, REM, SRED and DRED. Especially, NRED can stabilize queue length very fast even when a lot of flows suddenly crowd into the network. NRED is suitable to be deployed in today’s networks where hot spot frequently occurs.