前段时间看到brendangregg的 Linux Performance Analysis and Tools PPT里面提到的nicstat,研究了下是个不错的东西,分享给大家。

1
nicstat is to network interfaces as “iostat” is to disks, or “prstat” is to processes.

nicstat原本是Solaris平台下显示网卡流量的工具,Tim Cook将它移植到linux平台,官方网站见 这里。 相比netstat, 他有以下关键特性:

  • Reports bytes in & out as well as packets.
  • Normalizes these values to per-second rates.
  • Reports on all interfaces (while iterating)
  • Reports Utilization (rough calculation as of now)
  • Reports Saturation (also rough)
  • Prefixes statistics with the current time

我们来体验下,首先安装之,源码在 这里 下, 目前最新的版本是1.92。
解开后,由于这个版本默认是在32位linux下编译,所以需要改下Makefile.Linux:

1
2
3
4
5
6
7
$ diff Makefile.Linux64 Makefile.Linux
17c17
< CFLAGS = $(COPT) -m32
---
> CFLAGS = $(COPT)

$ sudo make -f Makefile.Linux install

就安装好了。

使用文档在这里: man nicstat
由于在linux下需要获取网卡的speed等信息,需要要以特权用户运行。

1
2
3
4
5
6
7
8
9
10
11
$ sudo enicstat -l
Int Loopback Mbit/s Duplex State
lo Yes - unkn up
eth0 No 1000 full up
bond0 No 0 unkn up

$ sudo enicstat
Time Int rKB/s wKB/s rPk/s wPk/s rAvs wAvs %Util Sat
13:04:10 lo 161.1 161.1 489.2 489.2 337.3 337.3 0.00 0.00
13:04:10 eth0 601.9 589.1 1868.1 1894.1 330.0 318.5 0.98 0.00
13:04:10 bond0 601.9 589.1 1868.1 1894.1 330.0 318.5 0.00 0.00

上面很特色的一个选项就是网卡的Util(利用率)以及Saturation(This the number of errors/second seen for the interface)在实践中用途挺大的。

通过strace和源码进行简单的分析:

1
2
3
4
open(“/proc/net/dev”, O_RDONLY) = 3
open(“/proc/net/snmp”, O_RDONLY) = 5
open(“/proc/net/netstat”, O_RDONLY) = 6
open(“/proc/uptime”, O_RDONLY) = 9

大部分统计信息是从上面3个文件获取到的,snmp提供了tcp和udp的统计信息。

默认统计信息 以KB为单位,-M选项可以更改为以M为单位。

1
2
3
4
5
$ sudo enicstat -M
Time Int rMbps wMbps rPk/s wPk/s rAvs wAvs %Util Sat
13:14:51 lo 1.26 1.26 489.1 489.1 337.3 337.3 0.00 0.00
13:14:51 eth0 4.70 4.60 1868.0 1894.0 330.0 318.5 0.98 0.00
13:14:51 bond0 4.70 4.60 1868.0 1894.0 330.0 318.5 0.00 0.00

nicstat提供了tcp链接外联和内联的个数,重置,Drops信息还有包重传率等信息,这些信息对于诊断tcp问题挺有用的。

1
2
3
$ sudo enicstat -t
13:09:40 InKB OutKB InSeg OutSeg Reset AttF %ReTX InConn OutCon Drops
TCP 0.00 0.00 2123.7 2251.7 1.84 0.79 0.000 7.16 2.95 0.00

最后需要注意的是, 解读输出信息的时候:

NOTES
On Linux, the NoCP, Defer, TCP InKB, and TCP OutKB statistics are always reported as zero.
The way that saturation is reported is a best effort, as there is no standardized naming to capture
all errors related to an interface’s inability to receive or transmit a packet. Monitoring %Util and
packet rates, along with an understanding of the specific NICs may be more useful in judging whether
you are nearing saturation.