Brent Simmons wants to see people move back to Web logs that render posts to static files so that they don’t go down in flames every time they get an unexpected traffic spike. I just wanted to point out that it is possible to build a resilient Web site using WordPress, as I explained in my post How to Speed Up WordPress In an Emergency.
I’ve never put a lot of faith in content management systems that “bake” pages if those pages will be updated dynamically. Obviously the worst case for any site is cratering under load, but the second worst is not showing people the most up to date content that’s available. If you allow comments on your site or provide other dynamic features, caching can be pretty tricky, and simply baking pages is almost completely impossible.
Fortunately, there are other ways to go about things. I still prefer caching at the database level to caching at the page level, but there’s no reason that a site that’s really dealing with a lot of traffic can’t do both.
March 18, 2011 at 5:29 pm
My 2c: Outsourcing comments to services that know what they’re doing wrt scaling takes care of that. Then (at least with Disqus) you just need static JavaScript to inject the comments into your page.
March 19, 2011 at 10:20 am
Or you can always include little bits of dynamic content using SSI, but keep the bulk of your website static. If even that becomes untenable under huge load, simply switch of server-side includes and you still have a functioning website.
March 19, 2011 at 11:41 am
I wouldn’t use SSI, but you could use AJAX comments or an IFRAME for the dynamic bits.
IIS has built-in page caching even for dynamic pages, with configurable TTLs etc.
An in-memory cache of most of the most-hit pages on a blog wouldn’t stretch any modern machine, and sending results right out of a string in memory is about as fast as it’s going to get.
The resilience argument is a good one although that’s more a question of “Why aren’t databases as robust as DNS or Apache?”
March 19, 2011 at 12:26 pm
The trouble with AJAX/iframe is that you don’t know if search engines will pick up on it. Though I guess some people’d consider that a feature rather than a problem, and don’t care about indexing comments too.
March 19, 2011 at 3:22 pm
I could say that for small sites, it doesn’t matter, but for big sites, baking is essential.
I work for a company whose site gets on the order of 2-3 million unique visitors/month. We regularly get Drudged, Slashdotted, Farked, Digged, Tumbl’d, etc., etc., etc. and it rarely causes disruption.
Baking our pages is the main thing that allows us to serve a heckofalot of content with limited (i.e., non-infinite) resources. We regularly update pages without issue, as our CMS takes care of rebaking everything that changes.
I had a chance to meet with Matt Mullenweg (the guy who invented WordPress) a few months ago and ask him about performance, etc. for WP. His suggestion was to have one master MySQL that handled writes + several slaves that handled reads. He also suggested throwing a bunch of commodity (i.e., cheap) app servers into a cluster. He actually said he didn’t like Supercache as it caused problems with concurrency.
It has been my experience that Movable Type is superior to WordPress if there’s a heavy load and one has limited horsepower. MT lost the war for non-technical reasons (6-apart basically dropped the ball: http://www.majordojo.com/2011/02/how-did-wordpress-win.php).
March 21, 2011 at 11:46 am
Just to be clear, Matt Mullenweg did not “invent” WordPress. He forked WordPress from a product called b2/cafelog. WordPress today is very different from the original version, and Matt gets credit for making it the success that it is.
The same can be said for MovableType. Ben Trott created MovableType but you can see a lot of the influence of a early blog solution called GreyMatter from a developer named Noah Grey in it.
March 21, 2011 at 11:53 am
Movable Type (even the original) and GreyMatter were two very, very different beasts. I don’t remember how alike they were in terms of user interface, but in terms of how they were written, they were not alike at all.
March 21, 2011 at 2:12 pm
Yes, I agree. They were different but I clearly saw the influence of GreyMatter within MT. I wish I had my old GreyMatter install so I could refresh my memory of how it worked. All I have is the HTML output as archived by archive.org for a site I built in GM back in 2001.
March 21, 2011 at 6:26 pm
Cam, I might still have a GM deployment stashed somewhere — if I find it, I’ll send you a copy. (I don’t remember what it looked like, but I switched to MT circa MT3, so I wouldn’t be able to compare it to early MT designs either.)
March 23, 2011 at 1:04 am
Thanks for the history lesson!