我尝试在同一网络中的 linux VM 上运行 tcp hping3，得到了 avg rtt ~5ms
sudo hping3 -S -p 22 10.1.0.8 -c 100
...
len=44 ip=10.1.0.8 ttl=64 DF id=0 sport=22 flags=SA seq=98 win=29200 rtt=0.6 ms
len=44 ip=10.1.0.8 ttl=64 DF id=0 sport=22 flags=SA seq=99 win=29200 rtt=1.4 ms
--- 10.1.0.8 hping statistic ---
100 packets transmitted, 100 packets received, 0% packet loss
round-trip min/avg/max = 0.6/5.2/9.7 ms
如果我使用 sockperf 工具测量延迟，则平均延迟约为 0.5 毫秒
(sockperf.bin ping-pong -i 10.1.0.8 -p 8302 -t 15 --pps=max )
sockperf output:
[[2;35m====> avg-lat=495.943 (std-dev=484.312)
[[0msockperf: # dropped messages = 0; # duplicated messages = 0; # out-of-order messages = 0sockperf: Summary: Latency is 495.943 usecsockperf:
[[2;35mTotal 15119 observations
[[0m; each percentile contains 151.19 observationssockperf: ---> <MAX> observation = 6839.398sockperf: ---> percentile 99.999 = 6839.398sockperf: ---> percentile 99.990 = 5292.623sockperf: ---> percentile 99.900 = 4023.327sockperf: ---> percentile 99.000 = 2434.115sockperf: ---> percentile 90.000 = 1005.612sockperf: ---> percentile 75.000 = 638.746sockperf: ---> percentile 50.000 = 360.516sockperf: ---> percentile 25.000 = 178.134sockperf: ---> <MIN> observation = 45.356
想知道这两个工具的延迟之间存在如此大差异的原因可能是什么。从 hping3 测量 tcp rtt 和从 sockperf 测量 tcp 延迟的内部方法是否相同？我在这里做错什么了吗？
为了验证我还尝试使用 psping 工具测量同一网络中两个 Windows VM 之间的 tcp 延迟
Connecting to 10.1.0.7:8888: from 10.1.0.6:62312: 1.03ms
TCP connect statistics for 10.1.0.7:8888:
Sent = 100, Received = 100, Lost = 0 (0% loss),
Minimum = 0.59ms, Maximum = 9.82ms, Average = 1.05ms