Chapter 7 - 16. Congestion Management in Ethernet Storage Networks以太网存储网络的拥塞管理

PFC and ECN

PFC is a hop-by-hop flow control mechanism. In contrast, ECN notifies the destination, which in turn informs the sender to reduce its traffic rate for congestion prevention. As previously explained, there is a delay between congestion detection (when the CE flag is marked) and when the reduced rate is observed by the congested switchport. This is approximately twice the round-trip time and the processing delay of the end devices. During this time, the queue on the congested switchport may fill up. Instead of dropping the packets, the hop-by-hop PFC may get activated, resulting in congestion spreading in the no-drop class. PFC 是一种逐跳流量控制机制。相比之下，ECN 会通知目的地，而目的地又会通知发送方降低流量速率以防止拥塞。如前所述，从检测到拥塞（标记 CE 标志时）到拥塞交换端口观察到降低的速率之间存在延迟。这大约是往返时间和终端设备处理延迟的两倍。在此期间，拥塞交换端口上的队列可能会填满。逐跳 PFC 可能会被激活，而不是丢弃数据包，从而导致拥塞在不丢弃类中扩散。

Using ECN and PFC together bring out the best of both. ECN is flow-based but its effect may be delayed. In contrast, PFC is priority- (class, or type) based but its prompt action avoids packet drops. Using them together makes the action prompt and flow based. 同时使用 ECN 和 PFC 可以发挥两者的优势。ECN 基于流量，但其效果可能会延迟。与此相反，PFC 基于优先级（类别或类型），但其及时行动可避免丢包。同时使用这两种方法，可以使操作既迅速又基于流量。

Configuring PFC and ECN Parameters

The rate-reduction action after notifying the end devices (by ECN and CNP) should eliminate the cause of congestion. When there is no congestion, there is no need to invoke PFC. Therefore, a properly working ECN should soon reduce the PFC Pauses. But configuring ECN thresholds on the switches require special consideration. The exact values depend on the switch type, so refer to the vendor documentation. This section explains only a conceptual overview. 通知终端设备（通过 ECN 和 CNP）后的速率降低操作应消除拥塞原因。没有拥塞时，就没有必要调用 PFC。因此，正常工作的 ECN 应能很快减少 PFC 暂停。但在交换机上配置 ECN 门限需要特别考虑。具体数值取决于交换机类型，请参考供应商文档。本节仅解释概念性概述。

Note the following points. 请注意以下几点。

1. Ingress queues/buffers start filling up only after egress queues are highly utilized. This is explained earlier in the section on Ingress and Egress Queues and Microburst Detection. 入口队列/缓冲区只有在出口队列被大量使用后才会开始填满。这一点在前面的 "入口和出口队列及微突发检测 "一节中已有解释。

2. The thresholds for ECN marking (such as by WRED) are applied on the egress queues, whereas the PFC Pause Threshold and Resume Threshold are applied on the ingress queues/buffers. ECN 标记（如 WRED）的阈值应用于出口队列，而 PFC 暂停阈值和恢复阈值则应用于入口队列/缓冲区。

3. Pause Threshold and Resume Threshold should be configured as per the details explained earlier in the section on Pause Threshold and Resume Threshold. Typically, for intra-datacenter links that are of short distance, changing the default Pause Threshold and Resume Threshold is not needed. 暂停阈值和恢复阈值应根据前面 "暂停阈值和恢复阈值 "一节中的详细说明进行配置。对于距离较短的数据中心内链路，通常不需要更改默认的暂停阈值和恢复阈值。

4. The ECN thresholds should be low enough to mark the CE flag much earlier to give enough time for the rate reduction action to be observed on the congested port. ECN 阈值应足够低，以便更早地标记 CE 标志，从而有足够的时间在拥塞端口上观察到速率降低操作。

5. The thresholds should be large enough to accommodate at least a few packets. For example, while enabling jumbo frames, a min size of 9000 bytes can’t even keep one full-size jumbo packet in the queue. 阈值应足够大，至少能容纳几个数据包。例如，在启用巨型帧时，9000 字节的最小大小甚至无法在队列中保留一个完整大小的巨型数据包。

Figure 7-18 to