Chapter 8 - 10. Congestion Management in TCP Storage Networks

本文主要是介绍Chapter 8 - 10. Congestion Management in TCP Storage Networks，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

Comparison with Lossless Networks

As Chapter 2 explains, Fibre Channel defines a maximum frame size of 2148 bytes. Users do not have to change the MTU value, and hence, it is never a consideration. 如第 2 章所述，光纤通道定义的最大帧大小为 2148 字节。用户无需更改 MTU 值，因此也无需考虑这个问题。

FCoE networks carry an FC frame within an Ethernet frame. Hence, the MTU should be increased to approximately 2300 bytes. FCoE 网络在以太网帧内传输 FC 帧。因此，MTU 应增加到约 2300 字节。

RoCEv2 networks have the same MTU considerations as TCP networks except for the MSS which is relevant only for TCP. The MTU on all the links in the end-to-end data path must be configured as per the maximum size of the packet send by the end devices to avoid fragmentation at the IP layer. RoCEv2 网络的 MTU 考虑因素与 TCP 网络相同，但 MSS 仅与 TCP 相关。端到端数据路径中所有链路的 MTU 必须根据终端设备发送的数据包最大大小进行配置，以避免在 IP 层出现分片。

Preventing Congestion in TCP Storage Networks

Unlike earlier chapters, this chapter explains congestion prevention before detection and troubleshooting mechanisms because TCP’s congestion control mechanisms are built in. Other mechanisms, like Explicit Congestion Notification (ECN), are standardized and well-adopted in the production networks. 与前几章不同的是，本章先解释拥塞预防，然后再解释检测和故障排除机制，因为 TCP 的拥塞控制机制是内置的。其他机制，如 "显式拥塞通知"（ECN），已经标准化，并在生产网络中得到广泛采用。

This book uses the term “congestion prevention” as an overall approach to eliminating or reducing congestion in a network so that application performance is acceptable. Do not confuse it with the “congestion control” and “congestion avoidance” mechanisms in TCP. These mechanisms slow down the sender to avoid packet drops, but the real cause of the problem must still be investigated and solved. 本书使用 "防止拥塞 "一词作为消除或减少网络拥塞的总体方法，以便使应用程序的性能可以接受。不要将其与 TCP 中的 "拥塞控制 "和 "拥塞避免 "机制混淆。这些机制会减慢发送者的速度以避免数据包丢弃，但问题的真正原因仍需调查和解决。