Build Scalable Web 2.0 Sites with Ubuntu, Symfony, and Lighttpd

One of the main concerns while developingThemBid.com was to have a website that would be easy to use but that could scale as the project grew. Fortunately, I’ve been writing web applications for a long time and along the way I’ve been exposed to issues ranging from modifying the Linux Kernel IP stack all the way to creating Web 2.0 Photoshop icons and CSS rounded corners. In this post, I will discuss the steps taken to optimize our initial ThemBid.com server to have a scalable website usingUbuntu,Lighttpd, PHP, andSymfony that can be applied to most Web 2.0 applications running on Linux. In the last part of this article I discuss an upgrade plan to scale your application as you grow.
Hardware - Initially we chose a decent server since we anticipated to get a few thousand users after the launch. The initial server is a dual core server with 2GB RAM and 2 x 36SCSI 10K RPM on RAID1. We decided to get this server fromLayeredtech since from past experience they’ve had a reliable network. Low-Latency DNS - Since the plan was to expand rapidly, something to consider was a low latency DNS server since we wanted our site to load fast while providing reliability and helping us with our future expansion and redundancy plan. We considered using our registrar’s DNS but they did not offerIP anycast so we decided to look elsewhere. We decided to go withDNS Made Easy since they were affordable and they had really low latencies. You can useDNS Stuff to test them. OS - Since Linux is widely supported in many hosting environments and proven to scale it was always assumed Linux was an option. Our programmers environment is a machine running Ubuntu with two monitors with Eclipse in 1.5 monitors, a terminal on the background, and a browser on half of the second monitor. Ubuntu was chosen as the OS on the server as well, and now with their extended software maintenance that Ubuntu offers in the server distribution, it was an easy choice. This is a screen shot of our development environment in case you are wondering.
Web Server - ThemBid.com has as many static pages/images as it does dynamic pages so we had to evaluate different servers. We initially looked at Apache with mod_php to server dynamic pages and a separate web server to serve static content. Then we took at look at Lighttpd. Lighttpd with FastCGI perfectly suits our needs. Lighttpd would handle all static image and “pass” the dynamic PHP page requests to FastCGI. Also, for those of you familiar with low-level event driven programming such as Kqueue, Epoll, etc. will know why an easy choice is Lighttpd. Lighttpd was written with speed in mind and they’ve done and excellent job. To install: # apt-get install lighttpd Then enable FastCGI on lighttpd with: # lighty-enable-mod fastcgi And edit /etc/lighttpd/conf-enabled/10-fastcgi.conf fastcgi.server = ( ".php" => ( ( "host" => "127.0.0.1", "bin-path" => "/usr/bin/php-cgi", "port" => 9000 ) ) ) Database - Since we have been developing MySQL based websites for a while and MySQL has scaled really good, we just decided this was not a time to test the waters again with PostgreSQL. MySQL slaves or a MySQL cluster would be enough for our scalability plan mentioned later. To install: # apt-get install mysql-client-5.0 mysql-common mysql-server mysql-server-5.0 Development Framework - The first framework we looked at was Rails since currently its one of the most talked about frameworks. The main two reasons why we didn’t use it is because we were already familiar with PHP and know it can scale such asDigg andYahoo and because its easier to find more experienced PHP programmers than Ruby programmers. After choosing PHP we were debating between usingSymfony orCakePHP. This was a tough choice, but at the time of making the decision Symfony had better documentation and seemed more active. In addition, at the same time Yahoo had released an application based on the Symfony framework and that helped us in our choice for Symfony. I am glad we made that decision!! To install PHP (some of these we will need later): # apt-get install php5-cgi php5-cli php5-common php5-gd php5-mysql php5-mysqli php5-sqlite php5-dev make Configure Lighttpd on /etc/lighttpd/lighttpd.conf to work with Symfony as the following assuming you install Symfony on /var/www/www.thembid.com $HTTP["host"] == "www.thembid.com" { server.document-root = "/var/www/www.thembid.com/web" url.rewrite-once = ( "^/(.*..+(?!html))$" => "$0", "^/(.*).(.*)" => "$0", "^/([^.]+)$" => "/index.php/$1", "^/$" => "/index.php" ) AJAX - If possible, AJAX should be used to improve the user experience while navigating the site. Another advantage this brings is that the number of requests potentially decreases since only the content needed to reflect an action on a page is retrieved. Symfony makes this tasks extremely easy with the help with the Javascript libraryPrototype. Caching - One of the most important things to make an application scalable is to do smart caching to prevent repetitive and redundant actions and instead use the available resources to perform high priority tasks ASAP. The following are some of the caches we introduced in our application. PHP Cache - By default, every time a PHP script is accessed, the web server must compile the script then execute the compiled code. This task is not needed if the PHP script doesn’t change. What a PHP accelerator does is it optimizes the scripts, compiles them, then caches them in a compiled state. Initially we triedXCache but we had an issue similar tothis in Symfony. So we decided to go foreAccelerator. To install it: # wget http://bart.eaccelerator.net/source/0.9.5/eaccelerator-0.9.5.tar.bz2 # tar xvjf eaccelerator-0.9.5.tar.bz2 # cd eaccelerator-0.9.5/ # phpize # ./configure # make install # mkdir /var/tmp/eaccelerator/; chmod 777 /var/tmp/eaccelerator/ add the following to /etc/php5/cgi/php.ini zend_extension="/usr/lib/php5/20051025/eaccelerator.so" eaccelerator.shm_size="0" eaccelerator.cache_dir="/var/tmp/eaccelerator" eaccelerator.enable="1" eaccelerator.optimizer="1" eaccelerator.check_mtime="1" eaccelerator.debug="0" eaccelerator.filter="" eaccelerator.shm_max="0" eaccelerator.shm_ttl="0" eaccelerator.shm_prune_period="0" eaccelerator.shm_only="0" eaccelerator.compress="1" eaccelerator.compress_level="9" then restart lighttpd Object and Content Cache - Object and content caching helps to eliminate the repetitive task of querying the database and re-computing the same result when our data hasn’t changed. Luckily Symfony already comes with multiple ways to implement caching. The main ones are file based caching, shared memory caching (using a PHP accelerator), andSQLite caching (we will release aMemcached based plugin soon). We anticipated to heavily use this cache so the number of items in the cache will be considerable. The file system cache was out of the question since we didn’t want our system spending all its time on file system operations trying to read and write these files. We wanted a cache that would solely live in memory for fast access. SQLite is really fast but it still resides on disk and the Memcached plugin was not written yet. We decided to use SQLite for its speed but put it on a memory based file system. This way we could get the speed on SQLite while sitting on memory. To do this we did the following: # mount -osize=100m -t tmpfs tmpfs /var/www/www.thembid.com/cache/ or you can do this from /etc/fstab tmpfs /var/www/www.thembid.com/cache/ tmpfs defaults,size=100m 0 0 Client Side Cache - Most of Javascripts, style sheets, and images would hardly change. As a result, we needed to tell the browsers not to re-download these for a while. Lighttpd comes with a module called mod_expire that can be configured to just do that. To configure edit /etc/lighttpd/lighttpd.conf and edit to make it look something like the following $HTTP["host"] == "www.thembid.com" { server.document-root = "/var/www/www.thembid.com/web" url.rewrite-once = ( "^/(.*..+(?!html))$" => "$0", "^/(.*).(.*)" => "$0", "^/([^.]+)$" => "/index.php/$1", "^/$" => "/index.php" ) expire.url = ( "/sf/" => "access 1 days", "/js/" => "access 1 days", "/css/" => "access 1 days", "/images/" => "access 3 days" ) }
Monitoring - In order to tell when it is time to upgrade the server, resources must be closely monitored. A quick and easy to do it is by usingMunin. To install: # apt-get install munin munin-node and change /etc/munin/munin.conf htmldir /var/www/www.thembid.com/web/status [thembid.com] address 127.0.0.1 use_node_name yes
Then visit yoursite.com/status and you will be able to closely monitor your resources. From there you can determine when and what to upgrade. Example:

In addition a web log analyzer such asAwstats should be installed to track hits and types of requests. This will let you know what parts of your site need to be emphasized the most for efficiency.
Scalability Plan - Depending on how the Munin graphs look, it will dictate the upgrade plan to scale your site. The first upgrade should be moving the MySQL operations to a new server. This will free up a lot of resources on this server including memory, disk, and CPU. After having more than one machine it makes sense to move to a Memcached distributed memory cache. If there is a database bottleneck a MySQL master/slave configuration can be added or a MySQL cluster if its affordable. If there is a need for more Lighttpd servers a load balancer such asLVS can be added in front to distribute the load.
As you can see, there are many enhancements that can be implemented to make a website scalable using Ubuntu, Lighttpd, MySQL, PHP, eAccelerated, SQLite, and Memcached while developing with Symfony. Keep in mind that this post does not include a way to make your servers fault tolerant. This will be the topic of my next post once we do that with ThemBid.com.