This is a simple throughput comparison of serveral TCP c/s implementations. I hope you can get a rough idea about their performance differences.
These toys do not measure the performance of receiving messages on the server side.
2019/01/31 15:40:42 payload size: 128, n: 1000000
2019/01/31 15:40:48 n: 1000000, time: 6.279, throughput: 159257
2019/01/31 15:39:18 payload size: 128, n: 1000000
2019/01/31 15:39:23 n: 1000000, time: 3.229, throughput: 309655
2019/01/31 15:40:50 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:41:02 n: 1000000, time: 11.878, throughput: 84189
2019/01/31 15:41:59 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:42:01 n: 1000000, time: 2.126, throughput: 470454
2019/01/31 15:39:25 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:39:35 n: 1000000, time: 6.176, throughput: 161907
2019/01/31 15:40:42 payload size: 128, n: 1000000
2019/01/31 15:40:48 n: 1000000, time: 6.279, throughput: 159257
2019/01/31 15:40:50 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:41:02 n: 1000000, time: 11.878, throughput: 84189
2019/01/31 15:44:35 payload size: 128, n: 1000000
2019/01/31 15:44:39 n: 1000000, time: 3.605, throughput: 277420
2019/01/31 15:44:41 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:44:47 n: 1000000, time: 6.001, throughput: 166645
2019/01/31 15:45:17 payload size: 128, n: 1000000
2019/01/31 15:45:22 n: 1000000, time: 4.486, throughput: 222919
2019/01/31 15:45:24 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:45:48 n: 1000000, time: 23.904, throughput: 41833
2019/01/31 15:46:27 payload size: 128, n: 1000000
2019/01/31 15:46:31 n: 1000000, time: 3.850, throughput: 259754
2019/01/31 15:46:33 payload size: 128, n: 1000000, TCP_NODELAY
2019/01/31 15:46:56 n: 1000000, time: 23.924, throughput: 41798
- Call IO functions as less as possible, no matter in what programming language.
- Sending messages in
gois about 50% percents slower than inc.