Chapter 7 - 17. Congestion Management in Ethernet Storage Networks以太网存储网络的拥塞管理

Congestion Management in VXLAN

As explained in the previous section, classifying and assigning the traffic to a no-drop queue preserves the lossless behavior of the traffic. This configuration must be consistent on all devices to preserve the end-to-end lossless behavior. 如上一节所述,将流量分类并分配到无损队列可保持流量的无损行为。这种配置必须在所有设备上保持一致,以保持端到端的无损行为。

Note the following points: 请注意以下几点:

1. Understanding Congestion: When PFC is enabled, congestion spreads in VXLAN as explained in the earlier sections. When an egress VTEP’s (or a leaf switch) queue starts filling up, it slows down the ingress traffic in the no-drop class by sending Pause frames. Consequently, the spine switch slows down all the traffic in that traffic class regardless of VXLAN encapsulation. Recall that for PFC, it does not matter how many headers are added to the packet. It just classifies and flow-controls the traffic using the DSCP field. The final state of congestion spreading is similar to as explained earlier in Figure 7-8. 了解拥塞: 启用 PFC 后,拥塞会在 VXLAN 中蔓延,如前几节所述。当出口 VTEP(或叶子交换机)的队列开始填满时,它会通过发送 "暂停 "帧来减缓不丢弃类中的入口流量。因此,骨干交换机会减慢该流量类中的所有流量,无论其是否采用 VXLAN 封装。回顾一下,对于 PFC 来说,数据包中添加多少报头并不重要。它只是使用 DSCP 字段对流量进行分类和流量控制。拥塞扩散的最终状态与图 7-8 中的解释类似。

2. Detecting Congestion: The congestion detection approach remains similar to what’s explained in the earlier section. The detection commands should account for DSCP-to-CoS mapping on the VTEPs. 检测拥塞: 拥塞检测方法与前一节中的解释类似。检测命令应考虑到 VTEP 上的 DSCP-CoS 映射。

3. Congestion Troubleshooting: The congestion troubleshooting methodology remains similar to what’s explained in the earlier section. While chasing the source of congestion, follow the traffic class on switchports or interfaces. Especially, on the spine switches do not be misled by the IP addresses because the VXLAN encapsulated packets have the IP addresses of the ingress and the egress VTEP in the outer header. Multiple flows (source-IP and destination-IP) are transported within the same VXLAN tunnel. Because of this reason, focus on monitoring traffic and Pause frames for no-drop traffic class instead of flows. 拥塞故障排除: 拥塞故障排除方法仍与前一节所述类似。在查找拥塞源的同时,要关注交换端口或接口上的流量类别。特别是在骨干交换机上,不要被 IP 地址误导,因为 VXLAN 封装数据包的外层报头中包含入口和出口 VTEP IP 地址。在同一 VXLAN 隧道中传输多个流量(源 IP

你可能感兴趣的:(Linux,kernel,网络,kernel,linux)