Correlation based optimization of random early detection

The performance of random early detection (RED) algorithm is known to be highly sensitive to network conditions and RED parameter settings. Under non-optimal settings the performance of RED tends to that of traditional tail-drop gateways. A black box parameter optimization scheme, with direct measurement of loss function, is proposed. We present a correlation based (CB) optimization algorithm, suitable for optimizing the RED parameters online. Network simulations are carried out to validate our approach. Simulation experiments show significant improvement in the performance of routers with negligible computational overhead.