Queuing determines how queued packets on an egress port are served. Each egress port supports eight queues, and three scheduling modes are available:
- Strict Scheduling: The queues are served in descending order (of queue number), so higher number queues receive higher priority. Queue7 has the highest priority, and queue0 has the lowest priority. The purpose of the strict scheduling mode is to provide lower latency service to higher classes of traffic. However, if the interface experiences congestion, the lower priority traffic could be starved.
- Simple Round Robin (RR): In round robin mode, the scheduler visits each backlogged queue, servicing a single packet from each queue before moving on to the next one. The purpose of round robin scheduling is to provide fair access to the egress port bandwidth.
- Weighted Round Robin (WRR): Each of the eight egress queues is assigned a weight value ranging from 0 to 63. The purpose of weighted round robin scheduling is to provide prioritized access to the egress port bandwidth, such that queues with higher weight get more of the bandwidth, but lower priority traffic is not starved.
A drop policy determines what happens when a queue is full or exceeds a minimum threshold. Depending on your switch model, you can select from one of two drop policies:
- The tail-drop drop policy is the default and is available on all platforms. When a queue is full, additional incoming packets are dropped until there is space available in the queue.
- The random early detection (RED) drop policy is available on 124D, 2xx, and 4xxD models. When the queue size exceeds the minimum threshold, packets are dropped at a constant rate until the queue is full. Using the RED drop policy helps improve the throughput during network congestion.
- The weighted random early detection (WRED) drop policy is an advanced version of RED and is available on 4xxE, 5xx, 1xxx, and 3xxx models. When the queue size exceeds the threshold, the WRED slope controls the rate at which packets are dropped until the queue is full. The drop rate increases when the queue buffer usage increases. If you select
weighted-random-early-detectionin the CLI, you can enable explicit congestion notification (ECN) marking to indicate that congestion is occurring without just dropping packets.