Refined design of random early detection gateways

Random early detection (RED) was proposed as an active gateway queue-management mechanism. This paper proposes to alter the RED design guideline that unconditionally allows transient congestion, and evaluates its impacts. This unconditional allowance of transient congestion is shown to be harmful when the queue is near full, because it causes buffer overflow at a gateway. Buffer overflow at a gateway leads to the global synchronization and oscillation of traffic load on the network. To effectively prevent buffer overflow at a gateway, the RED framework is refined in such a way that the gateway can detect a transient congestion in a timely manner and take actions to quench it when the queue is near full. Based on our simulation results, two enhancements are made in estimating average queue size. Equipped with these enhancements, the refined RED can strike a good balance between the allowance of transient bursty traffic and the avoidance of buffer overflow. Using extensive simulations, the refined RED is comparatively evaluated against and is shown to be superior to the original RED.