benches: add benchmark for concurrent connections

[niklasad1]

The code is a little bit !DRY but that's because the benchmark itself needs to construct the concrete client.

Interestingly, HTTP server throughput seems to be worse than the WS server for example if one compares one round trip with two connections it's around 15x slower on my machine.

When it comes to WebSockets the actual initial "connection" is excluded but it should at least show how efficient/slow polling each connections is... It's around 10x slower on my machine .

Concurrent connections with one roundtrip

sync/http_concurrent_connections/2
                        time:   [1.3427 ms 1.3654 ms 1.3898 ms]
                        change: [-2.9408% -1.1348% +0.6740%] (p = 0.24 > 0.05)
                        No change in performance detected.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild
sync/http_concurrent_connections/4
                        time:   [7.3308 ms 8.0426 ms 8.7470 ms]
                        change: [+159.46% +186.47% +212.97%] (p = 0.00 < 0.05)
                        Performance has regressed.
sync/http_concurrent_connections/8
                        time:   [24.833 ms 25.393 ms 26.038 ms]
                        change: [+388.94% +401.38% +414.71%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 11 outliers among 100 measurements (11.00%)
  4 (4.00%) high mild
  7 (7.00%) high severe
sync/http_concurrent_connections/16
                        time:   [9.3473 ms 9.4424 ms 9.5436 ms]
                        change: [-9.7526% -7.9180% -6.2010%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high severe
sync/http_concurrent_connections/32
                        time:   [18.828 ms 19.172 ms 19.545 ms]
                        change: [-8.1828% -5.7520% -3.2709%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  6 (6.00%) high mild
  1 (1.00%) high severe
sync/http_concurrent_connections/64
                        time:   [38.144 ms 38.908 ms 39.692 ms]
                        change: [-12.634% -9.7696% -6.9251%] (p = 0.00 < 0.05)
                        Performance has improved.

Roundtrip with one client

sync/http_round_trip    time:   [85.041 us 95.603 us 107.21 us]
                        change: [-48.697% -41.492% -32.649%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  4 (4.00%) high mild
  2 (2.00%) high severe

[maciejhirsz]

LGTM.

Out of curiosity, does the HTTP client preserve the connection to the server, or does it make a new connection on each request?

[niklasad1]

Out of curiosity, does the HTTP client preserve the connection to the server, or does it make a new connection on each request?

Good question, the answer is that it depends but we create the new clients in the bench setup (not included in benchmark) so it might or might not reuse the existing connection because the clients are used several times. If the client was only used for one request then I think we could be sure that the client didn't reuse an already existing connection.