Due to the high learning cost of the existing relatively complete network testing tools such as NS2 (network simulator), and the poor support for non-C / C + + written network protocols. The lightweight network test level tools, such as iperf3, do not give enough consideration to the horizontal expansion of supporting protocols. Now the latest version can only support TCP, UDP and STCP. However, there are hundreds of protocols in the application layer, and some developers have the need to test custom protocols. So iperf go is such a tool by implementing a very simple interface, it can measure the network speed of user-defined protocols.
The implementation refers to the C source code of iperf3 and is implemented in go language.
由于现有的较为完备的网络测试工具如 NS2 (Network Simulator) 学习成本较高，而且对于非 c/c++ 写的网络协议支持性不好。 而轻量级网络测试级工具例如 iperf3，本身架构上对支持协议的横向扩展考虑得不够到位，现在最新版也只能支持 TCP，UDP，STCP 三种协议，然而实际上应用层有上百种协议，并且部分开发者有测试自定义协议的需求。 所以 iperf-go 就是这样一个工具，通过实现极为简单的接口，就能够对自定义协议进行网络测速。
实现参考了iperf3的C的源码，用Go语言实现。
服务器端
./iperf-go -s
在默认参数下，在客户端启动对TCP测速
./iperf-go -c <server_ip_addr>
./iperf-go -c <server_ip_addr> -R // -R 表示server send, 测下行
服务器端
./iperf-go -s
在客户端启动对KCP测速
./iperf-go -c <server_ip_addr> -proto kcp
./iperf-go -c <server_ip_addr> -proto kcp -sw 512 -rw 512 // 设置发送接收窗口大小
更加详细的参数参考 ./iperf-go -h
> ./iperf-go -h
Usage of ./iperf-go:
-D no delay option
-P uint
The number of simultaneous connections (default 1)
-R reverse mode. client receive, server send
-b uint
bandwidth limit. (Mb/s)
-c string
client side (default "127.0.0.1")
-d uint
duration (s) (default 10)
-debug
debug mode
-f uint
flush interval for rudp (ms) (default 10)
-fr uint
rudp fast resend strategy. 0 indicate turn off fast resend
-h this help
-i uint
test interval (ms) (default 1000)
-info
info mode
-l uint
send/read block size (default 4096)
-nc
no congestion control or BBR (default true)
-p uint
connect/listen port (default 5201)
-proto string
protocol under test (default "tcp")
-rb uint
read buffer size (Kb) (default 4096)
-rw uint
rudp receive window size (default 512)
-s server side
-sw uint
rudp send window size (default 10)
-wb uint
write buffer size (Kb) (default 4096)
针对自己写的应用层协议，只需要实现几个简单的接口即可测速。
type protocol interface {
//name string
name() string
accept(test *iperf_test) (net.Conn, error)
listen(test *iperf_test) (net.Listener, error)
connect(test *iperf_test) (net.Conn, error)
send(test *iperf_stream) int
recv(test *iperf_stream) int
// init will be called before send/recv data
init(test *iperf_test) int
// teardown will be called before send/recv data
teardown(test *iperf_test) int
// stats_callback will be invoked intervally, please get some other statistics in this function
stats_callback(test *iperf_test, sp *iperf_stream, temp_result *iperf_interval_results) int
}
实现了这个接口之后，可获得最基本的带宽数据，对于 RTT，丢包或者其他自定义的统计变量，需要在 stats_callback 中从协议中获取
具体实现可参考已实现的 iperf_rudp.go
Server listening on 5201
Accept connection from client: 221.4.34.225:22567
[ ID] Interval Transfer Bandwidth RTT Retrans Retrans(%)
[ 0] 0.00-1.00 sec 1.50 MB 12.00 Mb/s 172.6ms 0 0.00%
[ 0] 1.00-2.00 sec 3.12 MB 25.00 Mb/s 169.1ms 0 0.00%
[ 0] 2.00-3.00 sec 2.75 MB 22.00 Mb/s 205.3ms 378 19.14%
[ 0] 3.00-4.00 sec 2.00 MB 16.00 Mb/s 164.8ms 1490 103.73%
[ 0] 4.00-5.00 sec 4.75 MB 38.00 Mb/s 162.9ms 873 25.59%
[ 0] 5.00-6.00 sec 1.25 MB 10.00 Mb/s 163.3ms 0 0.00%
[ 0] 6.00-7.00 sec 2.50 MB 20.00 Mb/s 163.3ms 0 0.00%
[ 0] 7.00-8.00 sec 2.62 MB 21.00 Mb/s 163.5ms 6 0.32%
[ 0] 8.00-9.00 sec 2.38 MB 19.00 Mb/s 162.7ms 0 0.00%
[ 0] 9.00-10.19 sec 2.50 MB 20.00 Mb/s 163.1ms 0 0.00%
- - - - - - - - - - - - - - - - SUMMARY - - - - - - - - - - - - - - - -
[ ID] Interval Transfer Bandwidth RTT Retrans Retrans(%)
[ 0] 0.00-10.19 sec 25.50 MB 20.40 Mb/s 169.1ms 2747 0.15% [SENDER]
