nginx+php-fpm性能参数优化原则

1.worker_processes 越大越好(一定数量后性能增加不明显)
2.worker_cpu_affinity 所有cpu平分worker_processes 要比每个worker_processes 都跨cpu分配性能要好;不考虑php的执行,测试结果worker_processes数量是cpu核数的2倍性能最优
3.unix domain socket(共享内存的方式)要比tcp网络端口配置性能要好
不考虑backlog,请求速度有量级的飞跃,但错误率超过50%
加上backlog,性能有10%左右提升
4.
调整nginx、php-fpm和内核的backlog(积压),connect() to unix:/tmp/php-fpm.socket
failed (11: Resource temporarily unavailable) while connecting to
upstream错误的返回会减少
nginx:
配置文件的server块
listen 80 default backlog=1024;
php-fpm:
配置文件的
listen.backlog = 2048
kernel参数:
/etc/sysctl.conf,不能低于上面的配置
net.ipv4.tcp_max_syn_backlog = 4096
net.core.netdev_max_backlog = 4096
5.增加单台服务器上的php-fpm的master实例,会增加fpm的处理能力,也能减少报错返回的几率
多实例启动方法,使用多个配置文件:
/usr/local/php/sbin/php-fpm -y /usr/local/php/etc/php-fpm.conf &
/usr/local/php/sbin/php-fpm -y /usr/local/php/etc/php-fpm1.conf &
nginx的fastcgi配置
   upstream phpbackend {
#    server  127.0.0.1:9000 weight=100 max_fails=10 fail_timeout=30;
#    server  127.0.0.1:9001 weight=100 max_fails=10 fail_timeout=30;
#    server  127.0.0.1:9002 weight=100 max_fails=10 fail_timeout=30;
#    server  127.0.0.1:9003 weight=100 max_fails=10 fail_timeout=30;
    server  unix:/var/www/php-fpm.sock weight=100 max_fails=10 fail_timeout=30;
    server  unix:/var/www/php-fpm1.sock weight=100 max_fails=10 fail_timeout=30;
    server  unix:/var/www/php-fpm2.sock weight=100 max_fails=10 fail_timeout=30;
    server  unix:/var/www/php-fpm3.sock weight=100 max_fails=10 fail_timeout=30;
#    server  unix:/var/www/php-fpm4.sock weight=100 max_fails=10 fail_timeout=30;
#    server  unix:/var/www/php-fpm5.sock weight=100 max_fails=10 fail_timeout=30;
#    server  unix:/var/www/php-fpm6.sock weight=100 max_fails=10 fail_timeout=30;
#    server  unix:/var/www/php-fpm7.sock weight=100 max_fails=10 fail_timeout=30;
   }
     location ~ \.php* {
       fastcgi_pass  phpbackend;
#      fastcgi_pass  unix:/var/www/php-fpm.sock;
       fastcgi_index index.php;
    ..........
    }
6.测试环境和结果
内存2G
swap2G
cpu 2核 Intel(R) Xeon(R) CPU E5405  @ 2.00GHz
采用ab远程访问测试,测试程序为php的字符串处理程序
1)
在开4个php-fpm实例,nginx 8个worker_processes 每个cpu4个worker_processes
,backlog为1024,php的backlog为2048,内核backlog为4096,采用unix domain
socket连接的情况下,其他保持参数不变
性能和错误率较为平衡,可接受,超过4个fpm实例,性能开始下降,错误率并没有明显下降
结论是fpm实例数,worker_processes数和cpu保持倍数关系,性能较高
影响性能和报错的参数为
php-fpm实例,nginx worker_processes数量,fpm的max_request,php的backlog,unix domain socket
10W请求,500并发无报错,1000并发报错率为0.9%
500并发:
Time taken for tests:  25 seconds avg.
Complete requests:    100000
Failed requests:     0
Write errors:      0
Requests per second:   4000 [#/sec] (mean) avg.
Time per request:    122.313 [ms] (mean)
Time per request:    0.245 [ms] (mean, across all concurrent requests)
Transfer rate:      800 [Kbytes/sec] received avg.
1000并发:
Time taken for tests:  25 seconds avg.
Complete requests:    100000
Failed requests:     524
  (Connect: 0, Length: 524, Exceptions: 0)
Write errors:      0
Non-2xx responses:    524
Requests per second:   3903.25 [#/sec] (mean)
Time per request:    256.197 [ms] (mean)
Time per request:    0.256 [ms] (mean, across all concurrent requests)
Transfer rate:      772.37 [Kbytes/sec] received
2)在其他参数不变,unix domain socket换为tcp网络端口连接,结果如下
500并发:
Concurrency Level:    500
Time taken for tests:  26.934431 seconds
Complete requests:    100000
Failed requests:     0
Write errors:      0
Requests per second:   3712.72 [#/sec] (mean)
Time per request:    134.672 [ms] (mean)
Time per request:    0.269 [ms] (mean, across all concurrent requests)
Transfer rate:      732.37 [Kbytes/sec] received
1000并发:
Concurrency Level:    1000
Time taken for tests:  28.385349 seconds
Complete requests:    100000
Failed requests:     0
Write errors:      0
Requests per second:   3522.94 [#/sec] (mean)
Time per request:    283.853 [ms] (mean)
Time per request:    0.284 [ms] (mean, across all concurrent requests)
Transfer rate:      694.94 [Kbytes/sec] received
与1)比较,有大约10%的性能下降
7. 5.16调整fpm的max_request参数为1000,并发1000报错返回降到200个以下,
Transfer rate在800左右
最后修改时间为:2015 年 05 月 13 日 11 时 17 分 AM
如果觉得我的文章对你有用,请随意赞赏

暂无评论

  1. onepx

    那么多并发用的什么工具呢?apache这边的ab只能用到cpu的一个核心,并发多了,ab占cpu为一个单核的100%,这时候不管服务器参数怎么调,ab这边的机器不行了啊...

    1. Memory
      @onepx

      @onepx:webbench http://www.lvtao.net/tool/webbench.html

  2. Memory

    可以继续调!max_request 主要是为了防止内存泄露用的,500很保守,因为并发访问量大是,每次进程回收都会导致很多的502,这是一个重要来源。fpm将max_request参数调整为2000, 报错率在0.2%以内,速度有所提升

发表评论