Choose the fastest php implementation

[waghanza]

Following #249 (comment), it'll be great to test php implementation

• Symfony
• Laravel

with new generation servers

• Swoole
• PHP-PM

Nginx / php-fpm was a default choice 😛

@bonfante if you are interested to see diffs between implementations, feel free to PR ❤️ we will have to choose the more performant / usable / standard

[doanguyen]

Hi,

An update from this issue: I had a micro-benchmark between Swoole Vs Nginx+Php-fpm. Most of the framework written in PHP now is sync so I have no way other than choosing vanilla PHP to perform the benchmark.

The benchmark workflow is like that:

Apache ab ---> Swoole
Apache ab ---> Nginx ---> PHP Application

I tried to set up Nginx as a reverse proxy on top of Swoole, but It does not work! Pain. Here is the result:

https://gist.github.com/doanguyen/4211d393a975ec361cd1f299182f9859

Conclusion:
Swoole, in fact, faster 16k > 13k fpm, but most of php library is blocking so it's too soon to have it on our stack, yet.

It seems php-pm is very good for managing php process, so I will take a look later.

[waghanza]

🎉 I'll be quite happy to see what's happening with those generation of tools

[waghanza]

@doanguyen you can also PR with https://github.com/php-pm/php-pm-docker

[waghanza]

@doanguyen I have tested php_pm here => https://github.com/waghanza/http-benchmark/blob/php_pm/php/laravel/Dockerfile

[doanguyen]

I've just made a quick test, it eats ram like twice as Chrome or even worse, workers don't seem to restart as they said.

[waghanza]

@doanguyen I've create an issue php-pm/php-pm-docker#18 to have some helps / see what's happening

[doanguyen]

I guess you miss that: https://github.com/php-pm/php-pm-httpkernel

setting up phppm is a pain.

[waghanza]

@doanguyen Nop, I use it https://github.com/waghanza/http-benchmark/blob/php_pm/php/laravel/composer.json

[doanguyen]

Did you try
ppm start --bootstrap=laravel --app-env=prod --logging=0 --debug=0 --workers=20
yet?

[doanguyen]

PHP-PM, 10 workers

$ wrk -c 10 -d 10s http://172.17.0.2:8080/user/0 -v --timeout 2s --latency
wrk 4.1.0 [epoll] Copyright (C) 2012 Will Glozer
Running 10s test @ http://172.17.0.2:8080/user/0
  2 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   193.17ms  284.81ms   1.20s    80.96%
    Req/Sec   209.65    130.33   670.00     68.60%
  Latency Distribution
     50%   12.97ms
     75%  339.05ms
     90%  692.05ms
     99%  978.06ms
  3675 requests in 10.03s, 1.90MB read
Requests/sec:    366.29
Transfer/sec:    193.74KB

with memory consumption

Not much better than: nginx+php-fpm

$ wrk -c 10 -d 10s http://172.17.0.2:8080/user/0 -v --timeout 2s --latency
wrk 4.1.0 [epoll] Copyright (C) 2012 Will Glozer
Running 10s test @ http://172.17.0.2:8080/user/0
  2 threads and 10 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    71.20ms  114.21ms 632.32ms   86.77%
    Req/Sec   173.41     81.91   363.00     61.70%
  Latency Distribution
     50%   21.52ms
     75%   39.95ms
     90%  241.27ms
     99%  507.64ms
  3280 requests in 10.01s, 1.79MB read
Requests/sec:    327.73
Transfer/sec:    183.01KB

and memory consumption

[doanguyen]

I think we stick to the nginx+php-fpm since php-pm is leaked in memory, and doesn't seem to fix in a short time.

[waghanza]

Souds good for me. Thanks for your work