Adaptive congestion avoidance for assured QoS

Traditionally, IP networks offer a single, best effort class-of-service. In order to support multiple services, the converged IP network must provide Quality of Service (QoS). One of the most difficult QoS areas is named as congestion avoidance. Today¿s router implementations depend solely on RED or lately WRED (Weighted Random Early Detection) for this function. However, two attributes minimize the effectiveness of WRED-only implementations. First, WRED simply determines congestion by looking at average buffer utilization. Second, dropping will be performed at the point of congestion instead of preventing the formation of a congestion point. To directly augment the short-comings of WRED, this paper develops an adaptive congestion avoidance algorithm, which prevents congestion within the IP core system, and assures deterministic QoS. It takes care of long-term congestion relief and ensures that a global precedence-based dropping is respected fairly among the ingress traffic.