Chapter 8 - 14. Congestion Management in TCP Storage Networks-CSDN博客

本文链接：https://blog.csdn.net/mounter625/article/details/136138544

高队列利用率可能导致高延迟，尤其影响存储阵列的I/O操作完成时间。网络端口的高利用率使得新数据包必须等待，增加了尾部延迟，影响应用性能。即使链路满载，仅监测链路利用率也难以察觉这些延迟。全闪存和NVMe存储设备的高性能要求关注队列管理和微爆发检测。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

Queue Utilization Considerations

TCP/IP networks typically aim for non-empty queues so that the queues always have packets for transmission without keeping the link idle. This achieves a high transmission rate and returns on investments (cost per Gbps). TCP/IP 网络通常以非空队列为目标，使队列中始终有数据包可供传输，而不会使链路处于空闲状态。这就实现了高传输速率和投资回报（每千兆比特成本）。

As Figure 8-12 shows, non-empty utilization can mean any of the other states—low, high, or even full utilization. In all these states, the link is transmitting at full capacity. But there are some side effects, especially when the queues are highly or fully utilized. 如图 8-12 所示，非空闲利用率可以指任何其他状态--低、高甚至满利用率。在所有这些状态下，链路都是在满负荷传输。但也有一些副作用，尤其是当队列被高度或完全利用时。

Latency

When queue utilization is high, although no packets are dropped, new packets must wait behind the already-queued packets. This significantly affects the completion times of the I/O operations. For example, on a 10 GbE port with 1500-byte frames, if the queue already has 100 packets, the next packet will be delayed by at least 120 microseconds. This is a long time for a packet to spend within a switch, while the All-flash and NVMe storage arrays can achieve I/O completion times in the order of 100s of microseconds. 当队列利用率较高时，虽然不会丢弃数据包，但新数据包必须在已排队的数据包后面等待。这会严重影响 I/O 操作的完成时间。例如，在具有 1500 字节帧的 10 GbE 端口上，如果队列中已经有 100 个数据包，则下一个数据包将至少延迟</