PHP7.1 is fast(?)

2017/?/? (forgot)
at somewhere.

this is kind of (joke|comedy).

Transcript

1. PHP7.1 is fast(?)
   

   View Slide

2. name:
   ͍ͩ͠͡ΎΜ͍ͪ
   (Ishida)
   nick:
   uzulla
   (͏ͣΒ)
   fun:
   PHP and !
   

   View Slide

3. conclusion
   7.1 is fast !
   (but any evidence here)
   

   View Slide

4. need benchmark !!
   

   View Slide

5. by the way...
   php can't to be standalone httpd. right?
   (php -s is for only development use)
   (btw, hhvm can be !)
   If will benchmark with apache/nginx, that is unfair(???)
   

   View Slide

6. use "ReactPHP"!
   (use libevent)
   standalone httpd(and other), that written by PHP!
   

   View Slide

7. Test Enviroment
   4 VPS (Vultr, $5/mo)
   4 1Core
   4 1GB mem
   4 ubuntu 16.10
   4 PHP7.1.2
   

   View Slide

8. Write git clone code.
   4 httpd, written by php
   4 https://github.com/uzulla/
   ReactPHP_PostSupportSample
   4 include monolog, twig(template engine.)
   

   View Slide

9. require_once 'vendor/autoload.php';
   $loader = new Twig_Loader_Filesystem(__DIR__.'/templates');
   $twig = new Twig_Environment($loader, [
   'cache' => __DIR__.'/twig_cache',
   ]);
   $access_log = new \Monolog\Logger('access');
   $access_log->pushHandler(new \Monolog\Handler\StreamHandler(__DIR__.'/access.log', \Monolog\Logger::INFO));
   // app
   $app = function (\Uzulla\React\Http\Request $req,\React\Http\Response $res) use ($twig, $access_log) {
   $path = $req->getPath();
   $method = $req->getMethod();
   $access_log->addInfo("{$method}\t{$path}");
   $params = [
   'params'=>$req->getParams(),
   'query'=>$req->getQuery()
   ];
   if($method === 'GET' && preg_match('|\A/\z|u', $path)){
   $res->writeHead(200, ['Content-Type' => 'text/html']);
   $res->end($twig->render('index.twig',$params));
   }elseif($method === 'POST' && preg_match('|\A/\z|u', $path)){
   $res->writeHead(200, ['Content-Type' => 'text/html']);
   $res->end($twig->render('index.twig',$params));
   }else{
   $res->writeHead(404, ['Content-Type' => 'text/html']);
   $res->end($twig->render('notfound.twig'));
   }
   };
   // build reactor
   $loop = React\EventLoop\Factory::create();
   $socket = new React\Socket\Server($loop);
   $http = new \Uzulla\React\Http\Server($socket);
   $http->on('request', $app);
   $socket->listen(8080);
   echo "running...\n";
   $loop->run();
   

   View Slide

10. $ git clone THAT
    $ composer install
    $ php boot.php
    listen...
    

    View Slide

11. ab(apache bench)
    4 ab -n 10000 -c 10 http://127.0.0.1:8080/
    

    View Slide

12. Requests per second:
    3477.39 #/sec
    

    View Slide

13. okay, try more concurrent connection.
    

    View Slide

14. 5000 concurrent
    4 ab -n 10000 -c 5000 http://127.0.0.1:8080/
    

    View Slide

15. fail
    4 socket: Too many open files (24)
    

    View Slide

16. ok, ab's limits.
    try another tool.
    apt install wrk
    4 https://github.com/wg/wrk
    

    View Slide

17. wrk
    4 wrk -c 5000 -d 5 http://127.0.0.1:8080/
    

    View Slide

18. Requests/sec
    98.16
    

    View Slide

19. tooooo slow,
    hmmm, concurrent overkill?
    try 1000 concurrent (same as ab's test).
    

    View Slide

20. 4 wrk -c 1000 -d 5 http://127.0.0.1:8080/
    4 Requests/sec: 83.17
    

    View Slide

21. !
    

    View Slide

22. okey, wrk is dosen't work.
    try with httperf !
    4 https://github.com/httperf/httperf
    

    View Slide

23. httperf
    4 httperf --port=8080 --server=127.0.0.1 --
    rate=2800 --num-conns=10000 --hog
    

    View Slide

24. Request rate:
    2799.6 req/s
    

    View Slide

25. seems good.
    but, is this correct?
    I need try more tools !
    

    View Slide

26. siege
    4 siege -c 100 http://127.0.0.1:8080/
    4 Transaction rate: 393.35 trans/sec
    

    View Slide

27. Untrustworthy...
    (Of course, these bench mark tools are more
    trustworthy than weird httpd.)
    

    View Slide

28. all results (?)
    ab: 3477.3 #/sec
    wrk: 98.1 req/sec
    httperf: 2799.6 req/s
    siege: 393.3 trans/sec
    

    View Slide

29. me:
    I'm screwed
    lestrrat ( a.k.a daisuke maki):
    Why don't you write a bechmarker yourself ?
    me:
    what?
    

    View Slide

30. me:
    OK, I'll write tools!
    (at YESTERDAY 11:36 pm)
    

    View Slide

31. I wrote "benchmark.php" !! (ta daaa!)
    

    View Slide

32. $c = 10000 ;
    $target = "http://127.0.0.1:8080" ;
    $start_usec = microtime(true);
    for( $i=0; $c>$i; $i++ ){
    $res = file_get_contents($target);
    }
    $used_time = (microtime(true) - $start_usec) ;
    echo ceil( 1/( $used_time/$c ) );
    echo "req/sec";
    

    View Slide

33. haha... yes, just kidding.
    let's play.
    

    View Slide

34. (unreliable) result
    4 3,731 req/sec
    

    View Slide

35. I can't believe this result.
    (faster than ab ;lol)
    it's not bad.
    but ...
    

    View Slide

36. I thought, this is slow benchmark tool. !
    I want more faster". but how?
    I need concurrent.!
    but, this is php.
    (curl_multi ? yes...but... )
    OK! WE NEED FORK! !
    

    View Slide

37. $max_worker = 10;
    $c = 10000;
    $c_ = $c/$max_worker;
    $target="http://127.0.0.1:8080";
    $start_usec = microtime(true);
    $pids=[];
    for($x=0;$max_worker>$x;$x++){
    $pid = pcntl_fork();
    if ($pid == -1) { die("fork failed");}
    else if ($pid) { $pids[]=$pid; }
    else {
    for($i=0;$c_>$i;$i++){ $res = file_get_contents($target); } exit;
    }
    }
    foreach($pids as $pid){ pcntl_waitpid($pid, $status); }
    $used_time = (microtime(true) - $start_usec) ;
    echo ceil( 1/( $used_time/$c ) ), "req/sec";
    

    View Slide

38. 2881req/sec
    (no fork ver 3,731 req/sec)
    

    View Slide

39. no no, more slow !!! !
    

    View Slide

40. !
    need more core!!!
    

    View Slide

41. create 4core 8GB mem server !
    

    View Slide

42. benchmark.php
    4147req/sec
    fork_benchmark.php
    3995req/sec
    

    View Slide

43. NO! why fork ver is more slow!? !
    

    View Slide

44. my httpd is BAD, that is why!!! !
    I use nginx to benchmark my benchmark!! !
    

    View Slide

45. result(?) with nginx
    4 benchmark.php
    4 6817req/sec
    4 fork_benchmark.php
    4 13730req/sec
    

    View Slide

46. OK!
    4 fork ver faster than not fork ver!!
    

    View Slide

47. ( what is ok? )
    

    View Slide

48. meanwhile in apache bench...
    4 ab -c 10 -n 10000 http://nginx/
    4 Requests per second: 15273.21
    

    View Slide

49. compare own tool and ab.
    4 fork_benchmark.php:
    4 13,730 req/sec
    4 ab:
    4 15,273.21 #/sec
    

    View Slide

50. okey!(?)
    

    View Slide

51. conclusion
    my bench (written by PHP) 0.9 times
    faster than AB !!
    ( DISCLAIMER, THIS RESULT IS REAL RESULT. BUT JUST
    PLAYING.)
    

    View Slide

52. OK !!
    

    View Slide

53. "what is ok?"
    (now 2:12 am)
    

    View Slide

54. Other benchmarks.
    4 php7.1.2 with my_httpd + fork_bench.php(c:10)
    4 3858req/sec
    4 php5.6.30 with my_httpd + fork_bench.php(c:10)
    4 2425req/sec
    4 hhvm3.18.1 with my_httpd + fork_bench.php(c:10)
    4 2377req/sec
    4 (but, hhvm w/o libevent)
    

    View Slide

55. conclusion
    PHP7.1 is fast
    (1.6x faster than 5.6 (at this test))
    

    View Slide

56. (REAL) conclusion
    USE RELIABLE TOOLS
    USE NGINX
    DON'T USE PHP FOR WRITE A
    BENCHMARK TOOL
    

    View Slide

57. but fun.
    PHP is waiting New Challengers .....
    FIN
    

    View Slide

58. //NEXT
    use pthread;
    

    View Slide