udp 的 GCP 虚拟机出口带宽限制
使用 N2 计算时，我们无法超越某些 PPS。无论我们使用哪种 N2 形状，数量都是恒定的。您能帮助我们了解导致问题的原因吗？
通常我们处理数据包大小为 32 字节、帧大小为 74 字节的 UDP(RTP) 流量。 当我们使用公共 IP 时，似乎发生了某种速率限制，并且我们的出口速度无法超过 115 Mbps。
对于相同数据包大小的专用 IP 流量，不会出现同样的问题。
we are not able to go beyond certain PPS while using N2 computes. The number is constant no matter which N2 shape we use. Can you please help us to understand what’s causing the problem?
Generally we deal with UDP(RTP) traffic of packet size 32 Bytes and frame size 74 Bytes.
When we use public IPs, it looks like that some kind of rate throttling is happening and we are not able egress beyond 115 Mbps.
Thee same issue is not observed with private IP traffic of same packet size.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（2）
Google Cloud通过限制输送到与VM相关的外部IP地址的入口流量来保护每个VM。限制为1,800,000 pps（每秒数据包） link 。
每个数据包的PPS限制 * 72个字节会导致127 mbits/sec，接近您观察到的极限。
Google Cloud protects each VM by limiting ingress traffic delivered to an external IP address associated with the VM. The limit is 1,800,000 pps (packets per second) link.
The PPS limit * 72 bytes per packet results in 127 Mbits/sec which is close to the limit you have observed.
几乎可以肯定的是，您遇到了外部/公共IP流量的明确限制，或者是同一固有的性能限制。根据2018年的“仙女座”的USENIX论文，该类型的流量与私人IP流量相比，通过网络/管理程序软件采取了不同的路径。
搜索饲料应该找到纸张/演示文稿，“ Usenix andromeda”。自2018年以来，当然可能至少发生了一些更改，但是其中所描述的内容的一般性应该广泛适用。
我建议您与官方支持渠道和/或您的帐户代表保持联系。
You are almost certainly running into either an explicit limitation for external/Public IP traffic or an inherent performance limitation for same. Per the 2018 Usenix paper for "Andromeda" that type of traffic takes a different path through the networking/hypervisor software than private IP traffic.
"usenix andromeda" as search fodder should find the paper/presentation. Of course there have likely been at least a few changes since 2018 but the general thrust of what is described therein should be broadly applicable.
I would recommend you get in touch with the official support channels and/or your account rep(s).