TCP拥塞控制

[WALL-E]

关键参数

• snd_cwnd 拥塞窗口
• snd_ssthresh 慢启动阈值
• snd_cwnd_clamp 拥塞窗口夹子

内核参数

• net.ipv4.tcp_allowed_congestion_control cubic reno

  Show/set the congestion control choices available to non-privileged processes. The list is a subset of those listed in tcp_available_congestion_control. Default is "reno" and the default setting (tcp_congestion_control).

• net.ipv4.tcp_available_congestion_control cubic reno

  Shows the available congestion control choices that are registered. More congestion control algorithms may be available as modules, but not loaded.

• net.ipv4.tcp_congestion_control cubic

  Set the congestion control algorithm to be used for new connections. The algorithm "reno" is always available, but additional choices may be available based on kernel configuration. Default is set as part of kernel configuration. For passive connections, the listener congestion control choice is inherited. [see setsockopt(listenfd, SOL_TCP, TCP_CONGESTION, "name" ...) ]

[WALL-E]

窥探连接

ss -ti sport == :22

State      Recv-Q Send-Q Local Address:Port                 Peer Address:Port
ESTAB      0      0      172.28.32.101:ssh                  172.28.66.22:63527
     cubic wscale:5,7 rto:201 rtt:0.888/0.845 ato:40 mss:1448 cwnd:10 ssthresh:136 send 130.5Mbps lastsnd:3 lastrcv:5 lastack:3 pacing_rate 260.8Mbps retrans:0/1 reordering:6 rcv_rtt:4 rcv_space:28960

几个关键值

• mss:1448
• cwnd: 10
• ssthresh: 136

按每条路由进行度量

[root@dockerHost-102 ~]# ip tcpmetrics|grep ss
23.23.78.154 age 541419.983sec rtt 349500us rttvar 349500us ssthresh 2 cwnd 1 metric_5 2796353 metric_6 1398176

自定义路由度量

ip route replace 172.28.32.102/32 via 172.28.32.101 ssthresh 24 initcwnd 10 initrwnd 65535 congctl reno

[WALL-E]

拥塞算法

Linux中不能关闭tcp拥塞算法。当前内核，默认的拥塞算法是cubic，可用的拥塞算法有：

• reno [Built-in]
• bic
• cubic
• westwood
• highspeed
• hybla
• htcp
• vegas
• veno
• scalable
• lp
• yeah
• illinois

设置方式

• sysctl sysctl -w net.ipv4.tcp_congestion_control = cubic
• setsockopt [see setsockopt(listenfd, SOL_TCP, TCP_CONGESTION, "name" ...) ]

查看内核支持哪些拥塞算法

[root@vagrant-172-28-32-101 linux-3.10.107]# cat /boot/config-`uname -r`|grep "CONFIG_TCP_CONG"
CONFIG_TCP_CONG_BIC=m
CONFIG_TCP_CONG_CUBIC=y
CONFIG_TCP_CONG_WESTWOOD=m
CONFIG_TCP_CONG_HTCP=m
CONFIG_TCP_CONG_HSTCP=m
CONFIG_TCP_CONG_HYBLA=m
CONFIG_TCP_CONG_VEGAS=m
CONFIG_TCP_CONG_SCALABLE=m
CONFIG_TCP_CONG_LP=m
CONFIG_TCP_CONG_VENO=m
CONFIG_TCP_CONG_YEAH=m
CONFIG_TCP_CONG_ILLINOIS=m
CONFIG_TCP_CONG_DCTCP=m

[WALL-E]

实验前置条件

1. 两台主机
   • 172.28.32.101 （服务端）
   • 172.28.32.102（客户端）
2. 安装sock程序(两个服务器) 参见 https://github.com/WALL-E/sock
3. 安装iproute2(172.28.32.102) 参见 https://github.com/WALL-E/static
4. 安装tcpdump
5. 前置条件, 关闭网卡offload特性（两个主机）

   ethtool -K eth0 tso off gso off gro off

6. 配置TCP拥塞控制参数（172.28.32.102）

   ip route replace 172.28.32.101/32 via 172.28.32.101 ssthresh 32 initcwnd 1 initrwnd 65535 congctl reno

[WALL-E]

吾生也有涯，而知也无涯。以有涯随无涯，殆已；已而为知者，殆而已矣！