search

dgzlopes / tcp-latency

Module and command-line tool to measure latency using TCP.
https://pypi.org/project/tcp-latency/
MIT License
58 stars 15 forks source link

RTT time is incorrect while using multithreading #18

Open s1042992 opened 3 years ago

s1042992 commented 3 years ago

In python, I created a thread list and append the 40 times ping works measure_latency(host, port) into the list, then the result was like:

CH. 3 =  15.075 ms
CH. 17 =  20.111 ms
CH. 16 =  20.111 ms
CH. 4 =  20.593 ms
CH. 9 =  20.111 ms
CH. 14 =  21.094 ms
CH. 6 =  20.595 ms
CH. 8 =  20.595 ms
CH. 7 =  24.604 ms
CH. 11 =  20.111 ms
CH. 13 =  24.606 ms
CH. 10 =  25.609 ms
CH. 12 =  20.094 ms
CH. 2 =  25.104 ms
CH. 15 =  25.107 ms
CH. 35 =  25.105 ms
CH. 32 =  29.118 ms
CH. 1 =  25.104 ms
CH. 29 =  29.117 ms
CH. 24 =  33.128 ms
CH. 5 =  35.132 ms
CH. 31 =  29.117 ms
CH. 20 =  34.632 ms
CH. 21 =  35.633 ms
CH. 22 =  33.628 ms
CH. 36 =  34.129 ms
CH. 30 =  38.139 ms
CH. 26 =  38.641 ms
CH. 18 =  38.641 ms
CH. 25 =  39.142 ms
CH. 33 =  36.134 ms
CH. 28 =  38.641 ms
CH. 40 =  43.153 ms
CH. 34 =  35.132 ms
CH. 37 =  44.158 ms
CH. 23 =  25.607 ms
CH. 38 =  42.652 ms
CH. 39 =  44.158 ms
CH. 19 =  34.632 ms
CH. 27 =  39.142 ms

The RTT time above somehow progressively increases.

But without multithreaded, the result will be like:

CH. 1 =  22.148 ms
CH. 2 =  10.528 ms
CH. 3 =  11.267 ms
CH. 4 =  20.77 ms
CH. 5 =  22.03 ms
CH. 6 =  20.608 ms
CH. 7 =  22.036 ms
CH. 8 =  21.07 ms
CH. 9 =  11.059 ms
CH. 10 =  10.388 ms
CH. 11 =  11.491 ms
CH. 12 =  21.23 ms
CH. 13 =  21.141 ms
CH. 14 =  10.883 ms
CH. 15 =  21.39 ms
CH. 16 =  10.012 ms
CH. 17 =  11.529 ms
CH. 18 =  22.938 ms
CH. 19 =  21.064 ms
CH. 20 =  10.528 ms
CH. 21 =  20.711 ms
CH. 22 =  21.616 ms
CH. 23 =  21.658 ms
CH. 24 =  20.516 ms
CH. 25 =  21.279 ms
CH. 26 =  10.48 ms
CH. 27 =  22.067 ms
CH. 28 =  21.048 ms
CH. 29 =  10.035 ms
CH. 30 =  21.028 ms
CH. 31 =  11.028 ms
CH. 32 =  21.891 ms
CH. 33 =  21.369 ms
CH. 34 =  10.487 ms
CH. 35 =  21.774 ms
CH. 36 =  20.57 ms
CH. 37 =  11.006 ms
CH. 38 =  10.755 ms
CH. 39 =  21.569 ms
CH. 40 =  10.053 ms

It seems very normal without multithreaded. I don't know why the result is incorrect in multithread, any help would be appreciated!

ATolkachev commented 2 years ago

@s1042992 it could be caused by GIL. Better to use multiprocessing instead of multithreading.