Wednesday, March 28, 2012

Varnish and Drupal

Varnish is front-end web proxy that I've been using with my Drupal sites for 18 months now. I talked about it in a presentation at Drupal camp with Khalid.

A front-end web proxy means that instead of visitors directly accessing the Drupal site, they go through a caching layer. I like to think of varnish as a protective bubble around Drupal.

Using Varnish has a number of benefits, and in my presentation I was particularly interested in demonstrating how it enables a site to survive huge spikes of sudden traffic, beyond what a web server would normally be able to handle. So that's one of the benefits.

Another important benefit is to reduce the load on the server, enabling better response. Or more cynically, it would enable you to pile more sites onto your server. But to explore this more closely: by moving most of your 'easy' page loads out of the php/apache stack, you can tune your stack better to what it's got to do - and you're not wasting 128Mb or more of RAM when serving an image, which is what happens by default on a server with apache and mod_php.

But even if you've got a pretty ordinary site and no server load issues, here's a reason why Varnish benefits you: your visitors' experience of speed. The key is that the traffic that is most sensitive to the load speed of your site is anonymous traffic to your front page. That's because new visitors (who usually end up first on your front page) are the ones you have to engage immediately or they go away (technically known as a "bounce"). A typical site might have a bounce rate of 50%, so any slowness in loading that front page by anonymous visitors is going to make a difference to whether that visitor continues to explore the site, or just leaves out of the tedium of waiting (who wants to wade through a slow loading site?). And here's why Varnish is the perfect solution: it does its best caching of your most visited pages to anonymous visitors!

And finally, some numbers. I was curious just how much traffic varnish might divert from a 'regular' drupal site, and haven't found anybody else's numbers.  Here are my last 2 months on one of my servers. The short answer is: about 30% of the visits go entirely to Varnish, more than half the hits (mainly because varnish caches images/css/js files for up to 2 weeks, even for logged-in users), and about half the total bandwidth.

Anybody got any comparable statistics?

 Apache:



MonthUnique visitorsNumber of visitsPagesHits
Feb 2012471448530546285476603111.79 GB
Mar 2012399636941748503083841612.76 GB


Varnish:


MonthUnique visitorsNumber of visitsPagesHits
Feb 201260082118672449906226430724.89 GB
Mar 20125171397369452394211089024.44 GB