Liveblog: Rasmus Lerdorf on Drupal Performance

Rasmus was born in Greenland, but lived for 8 years in Denmark, before moving to Canada in 1980 as a child. He says it’s a lot of fun coming back to Denmark.

He starts with a little history of PHP and his involvement in the web. In 1993, he says, it was basic HTML static pages. His manager wanted to have a catalogue online, and he didn’t want to do it all in static pages. He wanted to find a way to integrate web pages with databases.

The very first version of PHP, in 1994, had its code embedded in HTML comment tags. It used a database called miniSQL, a precursor to MySQL.

It got tiresome to work in HTML comments, so Rasmus read a ‘boring’ book on parsers, and he rewrote PHP in 1995 to be more like a language, and the PHP that we now know was born.

In 2003 there was a wish to move to more of an object oriented language. The every first implementation was ‘terrible’, as was the second. But now he believes they’ve done a decent job.

He says that now, in Drupal 7, “you guys have gone almost overboard on OO”!

Rasmus talks about some of the newer developments in PHP 5.3, including the new 5.3.3 release…

• Namespaces - using the “controversial backslash - get used to it, it’s not a big deal”
• goto - essentially very similar to multi-level breaks. The problem with multi-level breaks is when they are nested and you can’t work out where they end - but if you have a label and breakout of a loop to a label you know where you are. It’s not unrestricted.

Performance

Rasmus moves on to talk about performance. He says this isn’t just important for big sites, it matters for all sites:

• Latency puts users off, even on small sites
• Keep number of servers down
• Flexibility to add more features
• Pagerank takes performance into account
• Being slow is not cool

He checked out D7 a week ago and ran some benchmark tests against it. It achieved 8 requests per second on a small box with one core.

Rasmus says you should then enable APC or some other op-code cache - and be sure to set default timezone. The test site then achieved 31 requests per second. “Nothing makes as big a difference as op-code cache”.

Does the web server matter? Rasmus says no, it’s pretty much irrelevant whether you choose apache, nginx or anything else. He says it’s latency that matters and to set the max clients to around 40. If you need to handle more, then you need more servers. At Yahoo they set max clients to 25.

Rasmus says he’s going to repeat a rant he had at Drupalcon Szeged - why is Drupal obsessed with using require_once in code?
Starting with index.php it uses require_once to call bootstrap.inc - “why? It only gets called in one place. It doesn’t need to be a require_once”. He highlights a number of similar examples.
He says this creates an extra call, compared to a simple ‘include’ or ‘require’ - “Please, please clear this up”

He says the same about the DB autoloader. The database will be used on every page - why use an autoloader?

Rasmus is also not keen on the multiple settings files. Once the site is up and running you’re not going to change where the settings are stored. Yet Drupal checks round a lot of different directories to see if there are settings files. The conf path should just be hardcoded.

The original intention of PHP was to be a templating language - and then do heavy lifting in C. But C is hard so most people don’t do that. Why not look to the heaviest parts of code that don’t change often and create a C extension for Drupal? He identifies that drupal_render is one function that we need to focus on to get performance benefits.

HipHop

HipHop is Facebook’s method of compiling PHP into C code. Unfortunately D7’s use of PDO doesn’t work with HipHop’s PDO implementation and there are issues with the session array merging in includes/session.inc which you can hack around.

To sum up, Rasmus says:

• Performance is all about flexibility vs cost tradeoffs
• If you don’t know what things cost you lost
• Llearn tools, strace, callgrind, xdebug, xhprof
• check assumptions
• real performance is architecture driven

His slides are at http://talks.php.net/show/drupalconcph

Questions

Will being more OO attract more developers? Solving problems attracts people, beautiful code is a plus. It won’t attract people

What problems are there in the PHP dev process? People tend to solve problems they need. There may be a user need, but if the PHP development community doesn’t have that issue it doesn’t get done.

Why shouldn’t Drupal move away from PHP to Ruby or something else? End users don’t care what its written in. Switching Drupal away from PHP would take years, with no discernible improvements to users, and it’d be a huge waste of time

How do you keep yourself motivated after working on this so long? What keeps you excited about PHP. I’ve never been that excited about PHP. It’s easy to keep that level. PHP is a tool.