Tail drop is the default congestion avoidance mechanism on Cisco routers. Interface congestion occurs when a router receives packets faster than it can send them. When congestion exists, the excess packets are stored in a queue until the interface can transmit them. When the queue becomes full, the router drops all packets on the congested interface until there is room in the queue. This method of discarding packets is referred to as the tail drop mechanism. As the router drops packets, each sending device detects the packet loss and reduces its transmission rate, thereby reducing the congestion on the interface. This behavior is a feature of Transmission Control Protocol (TCP), which was designed to adjust transmit rates based on network conditions. However, because the tail drop mechanism does not differentiate between highpriority and lowpriority packets, all packets are dropped without regard to priority. Additionally, because the tail drop mechanism does not differentiate between packet flows, all TCP sessions are affected.

Global synchronization is a phenomenon associated with the tail drop mechanism. Because the tail drop mechanism discards packets for all TCP sessions, all sending devices reduce their transmission rates in unison. This behavior typically results in a lull in network traffic, causing the receiving router's interface to be underutilized. Then, as each TCP session attempts to maximize its transmission window, the interface rapidly becomes congested again, causing the router to tail drop packets. This cycle is referred to as global synchronization.

Queuing is a method of congestion management, not congestion avoidance. Every physical interface on a router has a hardware queue and a software queue. The hardware queue is always a first-in-first-out (FIFO) queue and has limited configuration options. The software queue can be configured for one of various queuing methods, such as weighted fair queuing (WFQ) or low latency queuing (LLQ).

TCP starvation is a phenomenon that occurs when TCP traffic is dominated by nonTCP traffic on an interface. Because nonTCP traffic, such as User Datagram Protocol (UDP) traffic, is not aware of packet loss due to congestion control mechanisms, devices sending nonTCP traffic might not reduce their transmission rates. This behavior causes the nonTCP traffic to dominate the queue and prevent TCP traffic from resuming a normal flow. To mitigate TCP starvation, you should avoid mixing TCP and UDP traffic in the same traffic class.