大网数据广告活动
服务器运维问答

What is an expected TCP retransmission rate

· 52CE 我爱测

I have a direct connection setup between my NAS and server with two gigabit links bonded in mode 0. I also have a single gigabit link from my server to my network. I want to test how the bond is performing in mode 0. netstat -s |…

问题原文

原帖链接 · Alec Fenichel · CC BY-SA 3.0

I have a direct connection setup between my NAS and server with two gigabit links bonded in mode 0. I also have a single gigabit link from my server to my network. I want to test how the bond is performing in mode 0.

netstat -s | grep 'segments retransmited' | awk '{print $1}' -> 21983

netstat -s | grep 'segments send out' | awk '{print $1}' -> 91874454

21983/91874454 = .000239272

Is this considered high? There is a lot of traffic between this server and the internet.

Also, is there a better way to test packet loss on just the direct connection between my NAS and server? ethtool stats are not available because the driver is not supported. Thanks.

已采纳解答

原帖链接 · Spooler · CC BY-SA 3.0

What is determined as "high" depends greatly on the application. In this case it's a fileserver using NFS, which is very forgiving to retransmissions.

In general if less than 1% of your traffic is redundant, you're in really good shape. You're far below that, so it doesn't look like this round-robin bond is causing retransmission issues (likely due to the fact that this is only a point-to-point link - a good use case for a mode 0 bond).

If you're having transmission problems regarding speed, you might look closer at jitter. This can be measured between two nodes that have iperf installed - not sure what your NAS is, though. balance-rr can cause issues with jitter, but in point-to-point links it shouldn't be a huge problem with NFS.

摘自公开归档,保留原文语言,版式经整理,外部媒体与邮箱地址已移除。

资料来源与适用范围

资料来源
Server Fault
适用范围
以原帖系统、软件版本与发布日期为准(2016-10-01)
内容语言
中文;命令、路径和错误标识保留技术原文
本地收录日期
2026-11-15
授权说明
CC BY-SA 3.0

正文与所需资源由本站本地存储。查看来源及更新记录 ↗

请按适用版本和实际日志确认原因;本站没有对每一种系统或设备逐项执行文中的操作。