Jeremy Smith's blog

Entry Is Labelled

Movable Type Scaling

There has been a little bit of a dust up regarding Movable Type and its ability to scale. I thought I would add to the conversation with my experiences.

I run the blog.case.edu system. It runs on a semi-modified installation of Movable Type. You can check on its web server stats here. There's some more stats here and here.

Here are some summaries of some of the actual numbers:

Approximately 50 GB/month is served over the wire on average (for February 2007, we're doing about 2.3GB/day for a total of 60.7GB so far this month)
446 entries were created in the last 30 days
5303 comments were made in the last 30 days (this number does include spam that got past the anti-spam measures), which is an average of 177 comments per day (the median number is 166 comments per day)

That would be a general summary of what we're serving here.

We haven't had to "scale" it. It runs on one Dell 1850 with nothing fancy going on -- no mod_(perl|FastCGI) or advanced caching mechanisms -- it's a MySQL install, Apache install, and the Movable Type CGIs1 all running on the Dell.

This isn't to say we've never had performance problems. It's a constant war with the spammers; one that is constantly fared with any large installation of weblogs or wikis. Sometimes we are on the losing edge of the battle, but that's mostly due to lack of engineering man-hour resources to implement all of the recommended anti-spam measures, and that blame should not be totally shouldered by MT. The CGIs aren't as responsive as I would like them to be, and we've been working to alleviate that problem. But overall, the performance of the system is good. I'm not saying there is not room for improvement, but in general, the system has held up.

In the original article questioning MT's ability to scale, they don't post any actual numbers of where its limitations begin to reveal themselves; so I'm not sure if the loads they are seeing are significantly higher than the numbers I'm talking about here. That may be the case.


1We are currently beginning the work on building the replacement for the blog.case.edu system, which will include performance enhancements such as running the CGIs in mod_fastCGI, running the comment and trackback handlers with a separate Apache instance, all new anti-spam counter-measures, and a host of other tweaks and improvements that we've been able to identify by running the system in production and analyzing server logs and load times. Our findings and tweakings may well deserve a separate blog entry.

Comments

  1. gravatar

    Actually, blog.case.edu (and its twin, wiki.case.edu) is a Dell PE 1750 (not much difference for the purpose of this discussion...same 3 GHz Xeon processors, slightly slower FSB & RAM...and sadly, old enough to have an expired warranty :-( ).

    Helpful Linux hint: you can find out a wealth of useful hardware info, including the model name and serial number (at least on Dell hardware) with: 'sudo dmidecode | less'. I believe that command will work on any Linux distro with reasonably modern BIOS & ACPI support (though I could be wrong).

Post a comment