Some non-scientific performance tests

[withinboredom]

I had fun looking at a production cluster running frankenPHP serving a zero-byte response (non-worker mode).

Running 3m test @ https://****/test
  44 threads and 1000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   153.46ms  161.06ms   6.05s    91.33%
    Req/Sec   158.06     37.99   560.00     68.29%
  Latency Distribution
     50%  115.99ms
     75%  204.36ms
     90%  301.28ms
     99%  554.76ms
  1208040 requests in 3.00m, 292.57MB read
Requests/sec:   6707.68
Transfer/sec:      1.62MB

^^ is an nginx ingress acting as TLS termination + proxy to frankphp.

Running 3m test @ https://****/test
  44 threads and 1000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     7.76s    11.72s   45.41s    83.79%
    Req/Sec    32.74     27.95   434.00     83.84%
  Latency Distribution
     50%  857.26ms
     75%   12.23s
     90%   29.75s
     99%   43.03s
  29146 requests in 3.00m, 9.65MB read
  Non-2xx or 3xx responses: 1066
Requests/sec:    161.84
Transfer/sec:     54.90KB

^^ approximately the same code (more headers and actually production) with the nginx ingress talking to php-fpm directly (in ondemand mode).

I have to say that I'm quite impressed with the performance of frankenPHP and I look forward to some actual scientific tests.

For the record, there are 3 servers in each case, with two instances of the server each.

• CPU: 12,12,16 cores
• RAM: 64Gb each