Building a Web Server 10x Faster Than Apache with Nginx 0.8.x + PHP 5.2.13 (FastCGI)

[Article Author: Zhang Yan | Version: v6.3

　 　Foreword: This is my sixth article on building an “Nginx + PHP (FastCGI)” web server. This series, one of the earliest domestic resources detailing the installation, configuration, and use of Nginx + PHP, has played a positive role in promoting Nginx’s adoption within China. This article may be updated with minor revisions; please bookmark the original link “http://blog.s135.com/nginx_php_v6/” for the latest content. The sixth article primarily introduces the new graceful restart method for Nginx 0.8.x, upgrades PHP to 5.2.14 with a PEAR fix, and upgrades the MySQL 5.1.x series to the 5.5.x series, which involves significant configuration file changes.

　　Links: 1st Edition, September 2007, 2nd Edition, December 2007, 3rd Edition, June 2008, 4th Edition, August 2008, 5th Edition, May 2009

　　Nginx ("engine x") is a high-performance HTTP and reverse proxy server, as well as an IMAP/POP3/SMTP proxy server. Nginx was developed by Igor Sysoev for Rambler.ru, Russia’s second most visited website, where it has been running for over three years. Igor released the source code under a BSD-like license.

　　Nginx’s superior performance and stability over Apache have led to its increasing adoption as a web server by domestic websites, including portal channels like Sina Blog, Sina Podcast, NetEase News, Tencent, Sohu Blog; video sharing sites like 6.cn, 56.com; well-known forums like the Discuz! Official Forum and SMTH Community; online game sites like Shanda Online and Kingsoft Xoyo; and emerging Web 2.0 sites like Douban, Renren, YUPOO Gallery, Kingsoft iCIBA, and Xunlei Online.

　　Official Chinese Nginx Wiki: http://wiki.nginx.org/NginxChs

　　Under high concurrent connections, Nginx is an excellent alternative to the Apache server. Nginx can also function as a Layer 7 load balancing server. According to my test results, Nginx 0.8.46 + PHP 5.2.14 (FastCGI) can handle over 30,000 concurrent connections, equivalent to 10 times that of Apache under identical conditions.

　 　Based on my experience, a server with 4GB of RAM running Apache (prefork mode) can typically handle only 3,000 concurrent connections, as these consume over 3GB of memory, leaving 1GB for the system. I once had two Apache servers where the MaxClients setting in the configuration file was 4000; when Apache’s concurrent connections reached 3800, the server’s memory and swap space were exhausted, leading to a crash.

　　Under 30,000 concurrent connections, this Nginx 0.8.46 + PHP 5.2.14 (FastCGI) server, with 10 Nginx processes active, consumed 150MB of memory (15MB*10=150MB). The 64 running php-cgi processes consumed 1280MB of memory (20MB*64=1280M). Combined with the system’s own memory usage, the total consumption was under 2GB. If the server has less memory, you can easily run only 25 php-cgi processes, reducing the total php-cgi memory consumption to just 500MB.

　　Under 30,000 concurrent connections, accessing PHP programs on the Nginx 0.8.46 + PHP 5.2.14 (FastCGI) server remained blazingly fast. The image below shows Nginx’s status monitoring page, displaying 28,457 active connections (details on configuring the Nginx monitoring page will be provided later in this article’s Nginx configuration file):

　　In my production environment, two Nginx + PHP5 (FastCGI) servers run multiple moderately complex pure PHP dynamic applications. A single Nginx + PHP5 (FastCGI) server’s processing capacity for PHP dynamic applications exceeded “700 requests per second“, equating to a daily capacity of 60 million visits (700*60*60*24=60,480,000) (more info here), while the server’s system load remained low:

　　On September 3, 2009, at 2:30 PM, the Kingsoft game “JX Online III” underwent a 1-hour emergency maintenance (http://kefu.xoyo.com/gonggao/jx3/2009-09-03/750438.shtml). A massive influx of players flocked to the official website. The dynamic application Nginx server cluster, handling forums, comments, and customer service, saw each server’s active Nginx connections reach 28,000. This was the highest concurrency value the author encountered with Nginx in a production environment.

　　Below is a stress test using 100 concurrent connections against two servers behind the same load balancer VIP providing identical services; one running Nginx, the other Apache. Nginx handled over twice the number of requests per second compared to Apache, with significantly lower system load and CPU usage:

　 　You can increase the connection count to 10,000~30,000 and stress test the phpinfo.php file on both Nginx and Apache. When accessing phpinfo.php on Nginx via a browser, everything functions normally. However, attempting to access the same file on the Apache server will result in a “page cannot be displayed” error. On a server with 4GB of RAM, even after optimization, Apache struggles to function normally under the stress of “webbench -c 30000 -t 60 http://xxx.xxx.xxx.xxx/phpinfo.php“, whereas a properly tuned Nginx can handle it.

　　webbench download link: http://blog.s135.com/post/288/

　　Note: When performing stress tests with webbench, the software itself consumes CPU and memory resources. For accurate testing, please install webbench on a separate server.

　　Test Results: ##### Nginx + PHP #####

　　Test Results: #####  Apache + PHP #####

　 　Why is Nginx’s performance so much higher than Apache’s? This is thanks to Nginx using the latest epoll (Linux 2.6 kernel) and kqueue (FreeBSD) network I/O models, whereas Apache uses the traditional select model. Currently, high-concurrency software on Linux like Squid and Memcached all use the epoll network I/O model.

　　For handling reads and writes of a large number of connections, the select network I/O model used by Apache is very inefficient. Let’s use an analogy to explain the difference between Apache’s select model and Nginx’s epoll model:

　 　Suppose you are studying at a university and live in a dormitory building with many rooms. Your friend comes to find you. The select-version dorm manager would take your friend door-to-door searching every room until they find you. The epoll-version dorm manager, however, would first record every student’s room number. When your friend comes, she just tells your friend which room you are in without having to personally lead your friend through the entire building searching. If 10,000 people come looking for their classmates living in this building, whose efficiency is higher — the select-version or the epoll-version dorm manager? The answer is self-evident. Likewise, in high-concurrency servers, polling I/O is one of the most time-consuming operations; the performance difference between select and epoll is equally clear.

　　Installation Steps:
　　(System Requirements: Linux 2.6+ kernel. The Linux OS used in this article is CentOS 5.3; also successfully installed on RedHat AS4)

　　1. Obtain Related Open-Source Programs:
　　1. [For CentOS] Use CentOS’s built-in yum command to install and upgrade required libraries (For RedHat and other Linux distributions, you can find the RPM packages for these libraries on the installation disc to install them):

　　2. [For RedHat] For RedHat and other Linux distributions, you can find the RPM packages for these libraries on the installation disc (check if required RPMs exist beforehand using commands like “rpm -qa | grep libjpeg”; typically the “xxx-devel” packages are missing and need installing). RedHat can directly use CentOS RPM packages. Here are the RPM download URLs:
　　①、RedHat AS4 & CentOS 4
　　http://mirrors.163.com/centos/4/os/i386/CentOS/RPMS/
　　http://mirrors.163.com/centos/4/os/x86_64/CentOS/RPMS/

　　②、RedHat AS5 & CentOS 5
　　http://mirrors.163.com/centos/5/os/i386/CentOS/
　　http://mirrors.163.com/centos/5/os/x86_64/CentOS/

　　③、RPM Package Search Sites
　　http://rpm.pbone.net/
　　http://www.rpmfind.net/

　　④、RedHat AS4 System Environment, typically missing support packages installation:
　　Ⅰ、i386 System

　　Ⅱ、x86_64 System

　　3. [Applicable to CentOS, RedHat, and other Linux operating systems] Download the program source packages:
　　All open-source software mentioned in this article are the latest stable versions as of July 26, 2010.
　　①. Download from the software’s official website:

　　②. Download from blog.s135.com (more stable; downloading the following software is only allowed from this site, or via commands like Wget or Curl under Linux/Unix):

　　II. Install PHP 5.2.14 (FastCGI Mode)
　　1. Compile and install the support libraries required for PHP 5.2.14:

　　2. Compile and Install MySQL 5.5.3-m3

　　Note: The following are additional steps. If you intend to run a MySQL database on this server, execute the steps below. If you only want PHP to support the MySQL extension library so it can connect to a MySQL database on another server, then the following two steps are not necessary.

　　①. Create the MySQL Database Storage Directory

　　②. Initialize the Data Tables as the MySQL User:

　　③. Create the my.cnf Configuration File:

　　Enter the following content:

　　④. Create a Shell Script to Manage the MySQL Database:

　　Enter the following content (the username ‘admin’ and password ‘12345678’ here will be created in the upcoming steps):

　　⑤、Grant the shell script executable permissions:

　　⑥、Start MySQL:

　　⑦、Log in to the MySQL server via the command line (press Enter when prompted for a password):

　　⑧、Enter the following SQL statements to create a user with root privileges (admin) and password (12345678):

　　⑨、(Optional) Stop MySQL:

　　3、Compile and Install PHP (FastCGI Mode)

　　4、Compile and Install PHP5 Extension Modules

　　5、Modify the php.ini File
　　Manual Modification: Find extension_dir = "./" in /usr/local/webserver/php/etc/php.ini
　　Change it to extension_dir = "/usr/local/webserver/php/lib/php/extensions/no-debug-non-zts-20060613/"
　　Add the following lines after this line, then save:
　　extension = "memcache.so"
　　extension = "pdo_mysql.so"
　　extension = "imagick.so"

　　Next, find output_buffering = Off
　　Change it to output_buffering = On

　　Then find ; cgi.fix_pathinfo=0
　　Change it to cgi.fix_pathinfo=0,Prevent Nginx File Type Mismatch Parsing Vulnerability.

　　Automated Modification: If you find manual editing tedious, you can run the following shell commands to automatically complete the modification of the php.ini file:

　　6. Configure eAccelerator to Accelerate PHP:

　　Press shift+g to jump to the very end of the configuration file, then add the following configuration information:

　　7. Create the www User and Group, and Directories for the Two Virtual Hosts blog.s135.com and www.s135.com:

　　8. Create the php-fpm Configuration File (php-fpm is a FastCGI management patch for PHP that allows for smooth changes to php.ini configuration without restarting php-cgi):
　　Create the php-fpm.conf file in the /usr/local/webserver/php/etc/ directory:

　 　Enter the following content (if you are installing Nginx + PHP for program debugging, please change 0 to 1 to display PHP error messages; otherwise, Nginx will report a blank 500 error page):

　　9、Start the php-cgi process, listening on port 9000 of 127.0.0.1, with 128 processes (if the server memory is less than 3GB, you can start only 64 processes), user www:

　 　Note: /usr/local/webserver/php/sbin/php-fpm has other parameters, including: start|stop|quit|restart|reload|logrotate. After modifying php.ini, you don’t need to restart php-cgi; use reload to reload the configuration file.

　　3. Install Nginx 0.8.46
　　1. Install the pcre library required by Nginx:

　　2. Install Nginx

　　3. Create Nginx Log Directory

　　4. Create Nginx Configuration File
　　①. Create the nginx.conf file in the /usr/local/webserver/nginx/conf/ directory:

　　Enter the following content:

　　②、Create the fcgi.conf file in the /usr/local/webserver/nginx/conf/ directory:

　　Enter the following content:

　　5、Start Nginx

　　IV. Configure Nginx + PHP to Start Automatically on Boot

　　Add the following content at the end:

　　V. Optimize Linux Kernel Parameters

　　Add the following content at the end:

　　Apply the configuration immediately:

　　VI. Smoothly Change Nginx Configuration Without Stopping the Service
　　1、After modifying the /usr/local/webserver/nginx/conf/nginx.conf configuration file, run the following command to check if the configuration file is correct:

　　If the screen displays the following two lines, it indicates the configuration file is correct:
　　the configuration file /usr/local/webserver/nginx/conf/nginx.conf syntax is ok
　　the configuration file /usr/local/webserver/nginx/conf/nginx.conf was tested successfully

　　2、Graceful restart:
　　①、For Nginx version 0.8.x, smoothly restarting the Nginx configuration is very simple now; just execute the following command:

　　②、For Nginx versions prior to 0.8.x, a graceful restart is slightly more involved. Follow the steps below. Enter the following command to check the Nginx master process ID:

　　The screen will display the Nginx master process ID, for example:
　　6302
　　At this point, execute the following command to apply the modified Nginx configuration file:

　　Or you can skip the hassle and locate the Nginx Pid file:

　　VII. Writing a Daily Scheduled Script to Rotate Nginx Logs
　　1. Create the script /usr/local/webserver/nginx/sbin/cut_nginx_log.sh

　　Enter the following content:

　　2. Set up crontab to rotate the Nginx access log daily at 00:00

　　Enter the following content:

　　Any minor updates to this article will be published first at the following URL:
　　http://blog.s135.com/nginx_php_v6/